Speak
成长迅猛的 AI 口语家教,需求信号真实,但信息披露不足,难以对独角兽估值拍板定价
Speak 的 AI 语言学习牵引力可信,也确有独角兽势头,但公开证据仍不足以支撑 $1B 估值。
封面要素
公司概况
Speak 是一家私营 AI 语言学习公司,专注于消费端与企业端的口语流利度培训,产品以语音优先为核心。公司将差异化的口语课程 体系与生成式 AI 及语音技术结合,产品验证比同类教育应用更扎实。公开信息显示全球分发规模显著、投资方阵容可信,但财务数据、 留存情况及股权结构表细节仍有太多空白,难以形成高置信度的定价判断。
- 成立时间
- 2016-01-01
- 创始人
- Connor Zwick, Andrew Hsu
- 创立地点
- San Francisco, California, USA
- 总部
- San Francisco, California, USA
- 产品
- Speak 通过移动应用及企业培训产品,提供 AI 引导的口语练习、发音反馈、结构化课程、实时角色扮演及多语言辅导。
- 客户
- 学习英语及其他语言口语的消费者,以及将英语技能提升与职场语言练习纳入员工福利的企业雇主。
- 商业模式
- 消费端应用内订阅,加上企业端按席位或合同制的语言培训服务。
- 阶段
- Series C private / growth-stage edtech unicorn
- 融资情况
- 2024 年 12 月完成 7,800 万美元 C 轮融资,估值 10 亿美元,累计融资约 1.62 亿美元;此前于 2024 年 6 月以 5 亿美元 估值完成 B 轮延伸融资。
执行摘要
主要优势
- 产品围绕口语流利度做差异化,而不是停留在被动语言学习。
- 消费端牵引力真实,应用商店评分强、下载量达数百万,国际覆盖也看得见。
- 2024 年公司证明投资人需求可以连续出现,估值六个月内从 $500M 翻到 $1.0B。
- Speak for Business 在消费者订阅之外,打开了第二个收入面。
- 工程和产品执行力看起来足以快速推出高级 AI 导师和语音智能体功能。
主要风险
- ARR、利润率、留存和回本周期的公开披露太薄,无法有把握地支撑当前估值。
- Duolingo、ELSA、Babbel、OpenAI 带动的新进入者以及低价 AI 导师都会施压,削弱定价权。
- 消费者计费摩擦、退款投诉和发音质量批评说明,增长叙事下面仍有产品风险边缘。
- 对第三方模型和应用商店生态的依赖,可能削弱护城河耐久度和经济性。
- 产品在多个地区处理语音数据,还可能触达低龄学习者,监管和隐私暴露随之上升。
未决问题
- 当前 ARR、毛利率、消费者转化漏斗和企业收入占比。
- Speak for Business 的续约、流失、净留存率(NRR)和集中度。
- Series C 估值下完整股权结构表条款、优先权、老股权利和投资人保护。
- 当前按职能划分的员工数、董事会构成和更深层治理控制。
- 分层队列留存,以及高价策略在早期采用者市场之外是否能持续。
目录
01公司概况
1.1 身份定位、产品边界与商业模式
Speak 是 Speakeasy Labs 的消费端品牌,总部位于旧金山,专注于口语流利度而非填鸭式背诵。无论是官网首页、应用 商店描述,还是 2024 年底的融资材料,公司始终将自己定位为推动学习者大声开口、即时获得反馈、建立对话自信的 AI 语言家教。这一定位比泛语言学习头部产品更聚焦——Speak 首先拼口语效果,而非宽泛的游戏化内容覆盖。 到 2026 年 5 月,Speak 的消费端公开产品已远超早期仅针对英语学习的切入口。官网和 Apple 应用页面列出六条 在售学习语言(法语、西班牙语、英语、韩语、意大利语、日语),2025 年 6 月的产品更新说明西班牙语率先面向英语 母语者推出,其后四门语言相继上线。这一演进支撑起混合商业模式:消费端订阅仍是核心变现引擎,但同一套语音优先 辅导技术栈如今也支撑着企业端雇主产品。TechCrunch 2024 年 12 月的报道给出了最清晰的消费端公开定价参考: 月付 $20 或年付 $99。 Speak 目前的信息披露仍是私营成长型公司的标准模式——公开内容足以确立品类定位、产品策略和收入规模概览,但 不足以验证单位经济模型、当前 ARR,或消费端与企业端的精确客户结构。[CO001, CO002, CO003, CO004, CO037]
1.2 创始人、管理团队与组织布局
公开的领导层画像是创始人主导、高度集中。2024 年底的融资报道一致点名 Connor Zwick 担任 CEO,Andrew Hsu 担任联合创始人兼 CTO;在官方公告和独立媒体报道中,Zwick 均是首要对外发言人。高度集中对决策速度和产品一致 性有利,但也形成明显的关键人依赖——公司的叙事、投资者沟通和产品愿景高度绑定于 Zwick 一人。 Speak 的运营版图对其规模而言相当国际化。TechCrunch 2024 年 6 月报道称公司拥有 75 名员工,分布在旧金山、 首尔、东京和卢布尔雅那;现有招聘页面仍标注首尔、卢布尔雅那和旧金山,将公司定性为国际分布式组织。不过招聘 页面内部存在信息滞后:其中还附有一段较旧的描述,提及 60 人团队和累计融资逾 6,000 万美元,说明 2024 年底 融资完成后页面并未全面更新。这是一个小但真实的尽调信号——即便是公司自控的公开界面,也应视为带时间戳的声明, 而非永久事实。 治理信息披露有限。公开信息中最可靠的董事会数据点是 Accel 合伙人 Ben Quazzo 在 C 轮后加入董事会。除此之外, 公开材料虽在 Form D 文件中列出关联人员,但并未披露完整的现任董事会名单、观察员权利或控制权条款。[CO002, CO014, CO015, CO016, CO024, CO029]
| 人物 | 职位 | 背景/公开信号 | 创始人-市场契合度或职能覆盖 | 关键人物依赖度 |
|---|---|---|---|---|
| Connor Zwick | CEO,联合创始人 | 官方帖子和 TechCrunch 融资报道中的主要外部发言人 | 掌控产品愿景、融资叙事和品类定位 | 高 |
| Andrew Hsu | CTO,联合创始人 | 见于融资报道,技术方向联合署名 | 掌控语音/AI 技术栈与技术公信力 | 高 |
| Ben Quazzo | Accel 合伙人,董事会成员 | 据官方及 TechCrunch 报道于 Series C 加入董事会 | 引入后期风险投资治理经验与招聘资源 | 中 |
| Colton Gyulay | Form D 关联人员 | 列于 Form D 关联人员名单 | 疑担任财务或法律领导职位,但公开范围不明 | 低 |
| Alex Berkenkamp | Form D 关联人员 | 列于 Form D 关联人员名单 | 疑担任运营领导职位,但公开范围不明 | 低 |
仅收录可直接见于已审查公开来源的职位;更广泛的管理层团队未完整披露。
[CO024, CO029, CO030]1.3 融资历史、估值跃升与文件记录时间线
Speak 的公开融资故事在 2024 年急速提速。2024 年 6 月,公司宣布完成 2,000 万美元 B-3 轮融资,估值 5 亿美元; 六个月后,C 轮融资 7,800 万美元,估值翻倍至 10 亿美元。C 轮公告称累计融资达 1.62 亿美元,并将本轮定性为 年内第二笔被抢先推进的融资。TechCrunch、Tech Funding News 和 HolonIQ 的独立报道充分印证了估值跃升, 独角兽里程碑视为有据可查。 证券备案记录与融资叙事在方向上吻合。2024 年 12 月 11 日提交的 Form D 显示一笔 7,770 万美元的 506(b) 规则股权发行,首次出售日期为 2024 年 11 月 13 日。备案记录还显示 2024 年 8 月 12 日和 2023 年 10 月 17 日各有一次 Form D 事件,可能对应 B 轮延伸期间的中间融资活动及此前的私募股权发行。这些备案未披露持股 条款、优先权或全部治理权利,但它们提供了有用的一手文件佐证——真实资金在公司所述的时间线上确实发生了流动。 此后,HolonIQ 的 EdTech 独角兽名册于 2024 年 12 月将 Speak 列入全球 EdTech 独角兽榜单,估值 10 亿美元。 C 轮由此不仅是融资里程碑,更是品类地位里程碑,对未来融资、人才招募和企业销售具有真实信号价值。[CO008, CO009, CO010, CO011, CO012, CO023]
| 利益相关方 | 角色 | 控制权或经济重要性 | 尽调问题 |
|---|---|---|---|
| Accel | Series C 领投方;通过 Ben Quazzo 持有董事席位 | 锚定独角兽轮,2024 年 12 月起可能持有重要治理权 | 需获取董事会文件、优先认购权和持股比例。 |
| OpenAI Startup Fund | 持续跟投方 | AI 生态战略信号,兼任产品验证合作伙伴 | 需明确任何模型访问、品牌或排他性安排。 |
| Khosla Ventures | 持续跟投方 | 多轮长期财务支持方 | 需确认持股比例、优先权条款和跟投意向。 |
| Y Combinator / Paul Graham 人脉网络 | 早期投资人,信号放大器 | 高声誉价值与创始人网络支持 | 需明确 YC 权利是否与标准种子投资人不同。 |
| Buckley Ventures | 主导 2024 年 6 月 Series B-3 | Series C 跃升前的关键过桥投资方 | 需审查 B-3 是否在 Accel 领投 Series C 之前引入特殊条款。 |
| 使用 Speak for Business 的雇主 | 商业利益相关方而非股权持有人 | 若 200+ 客户声明属实,可能是有价值的设计合作伙伴 | 需获取按席位数、地区和续约状态排序的前 10 大客户。 |
公开来源识别了主要融资参与方和企业客户类别,但未披露完整股权结构表或持股比例。
[CO008, CO009, CO011, CO017, CO018, CO023]| 日期 | 事件 | 类型 | 金额/估值/状态 | 参与方 | 意义 |
|---|---|---|---|---|---|
| 2016 | Speak 成立/正式创建公司进入公开记录 | 创立 | Connor Zwick; Andrew Hsu | 确立了官方及第三方资料使用的公司主体身份。 | |
| 2019 | 英语学习首先在韩国上线,开启亚洲市场楔形策略 | 产品 | 初始市场发布 | Speak | 确立了韩国优先的分发策略,后被官方帖子多次援引。 |
| 2023-10 | 公开文件记录显示一份日期为 2023-10-17 的 Form D | 融资 | Form D 事件 | Speakeasy Labs | 本次审查中最早的私募资本活动文件证据。 |
| 2024-06-18 | Series B-3 宣布 | 融资 | $20M,估值 $500M | Buckley Ventures、OpenAI Startup Fund、Khosla、Paul Graham、Jeff Weiner 等 | 估值不到一年翻倍,为独角兽跃升奠基。 |
| 2024-08-12 | 公开文件列表中出现额外 Form D | 融资 | Form D 事件 | Speakeasy Labs | 提示 Series C 交割前存在进一步私募发行活动。 |
| 2024-12-10 | Series C 宣布;Speak for Business 获重点展示;Ben Quazzo 加入董事会 | 融资/治理 | $78M,估值 $1B | Accel、OpenAI Startup Fund、Khosla、YC 等 | 独角兽里程碑,治理升级,董事会正式化。 |
| 2024-12 | HolonIQ 将 Speak 纳入教育科技独角兽榜单 | 规模 | $1B 估值状态 | HolonIQ | 超出公司公关范围的外部品类认可。 |
| 2025-06-17 | 面向英语使用者的四种新语言上线,学习者数量突破 15M+ | 产品/规模 | 声称学习者 15M+ | Speak | 确认多语种扩张超越原有英语楔形策略。 |
| 2026-02-23 | Android Police 发布关于发音宽松度的批评性评测 | 负面 | 独立负面产品评测 | Android Police | 反映出规模化曝光,同时也暴露了过度宽松反馈的真实质量风险。 |
尽管有一行 TechCrunch 记录提及 2014 年启动,创立年份以 2016 年为准;该矛盾保留于证据缺口,而非在正式记录的历史时间线中留作悬案。
[CO001, CO008, CO009, CO012, CO021, CO023]2016 年至 2026 年 2 月,Speak 身份定位、资本形成、产品扩张和外部认可的公开年表。
[CO008, CO009]截至 2026 年 5 月,Speak 公开成熟度、吸引力和尽调风险的简明视图。
[CO012, CO032]1.4 牵引力信号、企业端扩张与风险标志
Speak 的公开牵引力数据足以支撑"规模化私营公司"的判断,但不足以核准运营表现。消费端方面,官方渠道声称 1,500 万次以上下载量和 4.8 分评分,应用商店独立验证了大量评分深度(Apple 44,000 条评分;Google Play 112,000 条评论)。2024 年初的融资材料称 Speak 在 40 多个国家拥有逾 1,000 万学员,当年用户开口说话已超 10 亿句。这些均为公司自报指标,但与产品可见的分发规模基本吻合。 最值得关注的非消费端信号是企业产品。Speak for Business 看似在 2024 年 6 月至 12 月两轮融资之间正式起步, C 轮公告声称企业客户已超 200 家、员工使用率达 85%。当前 B2B 官网同样表示超过 200 个品牌依赖 Speak for Business。这表明订阅之外确实存在真实的第二增长线,尽管公开信息仍未拆分 B2B 收入、合同规模 或留存率。 风险标志总体与产品质量和信息披露有关,而非法律或财务层面。独立评测普遍称赞产品打磨程度和口语优先设计, 但负面评价也指出发音评分过于宽松、高级套餐包装混乱、退款及自动续费投诉,以及员工数和董事会构成的信息披露 不完整。这些问题不能否定增长叙事,但也说明下一层尽调工作将偏向运营细节,而非纯粹的宣传审核。[CO005, CO006, CO007, CO013, CO017, CO018]
| 指标 | 数值/状态 | 日期 | 置信度 | 缺口/尽调问题 |
|---|---|---|---|---|
| 最新估值 | $1.0B(Series C) | 2024-12 | high | 有官方公告、TechCrunch 和 HolonIQ 支撑。 |
| 累计融资 | $162M 累计 | 2024-12 | medium | 公司口径;需与完整股权结构表及任何未披露过桥资本核对。 |
| 消费者规模 | 声称下载量 15M+;2024 年中报道用户超 10M | 2025-06 / 2024-06 | medium | 下载量与用户数分母不同;需获取 MAU、DAU 和付费订阅用户数。 |
| 企业牵引力 | 声称 200+ 客户/品牌;采用率 85% | 2024-12 / 2026-05 | medium | 需客户名单、席位数和续约数据以验证企业深度。 |
| 当前员工数 | 2024 年 6 月 75 人;2025 年第三方估算 253 人 | 2024-06 / 2025-11 | low | 需按职能获取当前员工数;公开信息存在矛盾。 |
| 消费者标价 | $20/月或 $99/年 | 2024-12 | high | 需按地区核实当前应用内定价及 Premium Plus 增购规则。 |
| 应用口碑 | Apple 4.8;Google Play 4.7 | 2026-05 | high | 需按地区和版本查阅低分投诉群体。 |
| 债务/信贷额度 | low | 已审查来源中未发现公开债务或信贷安排;需直接与管理层确认。 | ||
| 董事会构成 | Series C 增加 Ben Quazzo;完整董事会未公开 | 2024-12 | low | 需获取当前董事会名单、观察员权利和治理文件。 |
不受支持的私有指标保留为 null 或描述为相互矛盾的公开估算,而非强行填入虚假精确数字。
[CO005, CO006, CO007, CO009, CO010, CO018]Speak 核心 AI 辅导、订阅模式、企业扩张和融资势头之间的连接关系。
[CO003, CO009, CO017, CO018, CO019, CO032]1.5 展示图表
02市场分析
2.1 市场边界:Speak 参与竞争的赛道与未参与的赛道
Speak 所在的细分市场比泛"语言学习"更窄,可界定为数字化、口语优先的语言学习:借助 AI 或真人对话帮助 用户在真实语境中练习口语输出,而非以阅读、语法或单词卡片记忆为主。Speak 的融资材料明确了这一框架—— 将口语优先家教与旧式应用模式对比,并将那个更宽泛的 1,000 亿美元以上线上线下语言学习市场仅作为背景信息。 这意味着最近的替代品不是所有教育应用,而是一批更窄的竞品:Duolingo、Babbel、Busuu 等宽泛语言平台, ELSA、Praktika 等专注口语的应用,Cambly 等真人家教网络,以及企业场景下的雇主赞助英语培训或内部辅导。 属于该市场的消费包括:消费端订阅、雇主出资的英语技能提升,以及数字化口语类产品的支出。市场外的消费 包括:应试备考、不以口语为重点的教材主导课堂教学,以及与语言流利度无关的企业培训类别。 关键尽调结论是:Speak 最好被理解为嵌入庞大但分散市场中的口语流利度楔子。总可用市场(TAM)足够大,但 可核准的市场规模取决于谁真正愿意为反复口语练习付费,以及多大比例的需求可以数字化。[CM001, CM002, CM023, CM030, CM031, CM033]
| 细分/品类 | 纳入支出 | 排除支出 | 买方/付款方 | 对 Speak 的相关性 |
|---|---|---|---|---|
| AI 口语优先语言应用 | 消费者订阅与雇主资助的口语练习 | 通用课堂学费及无关职场培训 | 学习者或雇主 | 核心市场 |
| 更广泛的数字英语学习 | 应用、在线辅导、应试练习、数字英语课程 | 无数字成分的纯线下课堂教学 | 学习者、家长、雇主、学校 | 重要的邻近市场池 |
| 广义在线语言学习 | 跨多语种的所有在线语言学习订阅 | 线下辅导与教材销售 | 学习者、雇主、学校 | 上界参考,对 SAM 过于宽泛 |
| 人类辅导与实时对话服务 | 实时课程与循环辅导会员 | 自助 AI 纯订阅 | 学习者或雇主 | 主要替代品,非直接 AI 市场 |
| 社区/结构化课程应用 | 含阅读、语法和社区纠错的订阅课程 | 纯口语或纯辅导模式 | 学习者 | 消费者预算的主要替代选项集 |
边界逻辑将口语流利度产品与更广泛的教育和辅导领域区分开来。
[CM001, CM023, CM030, CM031, CM033]全球英语学习需求如何转化为消费者订阅或雇主付费采纳。
[CM004, CM023]2.2 市场规模测算:宏观需求真实存在,但各口径不可互换
最有力的公开证据支持一个庞大且持续增长的数字英语学习市场,但不存在单一的清晰数字。Technavio 估计 2024 至 2029 年数字英语学习市场将增长 394.6 亿美元,年复合增长率 24.5%,亚太地区贡献 39% 的增量。 Preply 采用不同方法和更宽泛的在线学习口径,估计 2025 年全球在线语言学习市场约达 1,150 亿美元,其中 仅英语市场约为 435.1 亿美元,年增速约 22%。Speak 自己在 2024 年 6 月使用了最宽泛的口径——描述为 1,000 亿美元以上的线上线下市场。 这些数据不能取平均。Technavio 报告的是特定数字英语细分市场的预测增量;Preply 报告的是平台驱动的更 宽泛在线市场;Speak 则使用了包含线下支出的公司自撰框架。正确的解读是方向性而非精确性:所有可信来源 均指向一个巨大且仍在增长的市场,但公开数据无法为 Speak 提取出经过来源验证的清晰可服务市场(SAM)或 可获取市场(SOM)。 更具操作价值的规模判断是定性的:英语仍是最大的单一语言学习品类,亚太仍是最重要的增长地域,口语优先 的数字产品正在乘着英语长期需求与 AI 辅助学习短期采用的双重东风前行。[CM001, CM003, CM004, CM006, CM007, CM010]
| 来源/视角 | 年份 | 地区 | 数值 | 方法论 | 置信度 | 局限性 |
|---|---|---|---|---|---|---|
| Technavio 数字英语市场增长 | 2024-2029 | 全球 | 增长 USD 39.46B;CAGR 24.5% | 数字英语学习细分市场预测增长 | medium | 增量而非当前市场规模 |
| Technavio 亚太贡献 | 2024-2029 | 亚太 | 占增长的 39% | 预测市场增长的区域贡献 | medium | 反映地区分布,非 Speak 份额 |
| Preply 在线语言学习市场 | 2025 | 全球 | USD 115B | 平台主导报告,采用更广泛的在线语言学习视角 | low | 竞争对手撰写,范围宽于 Speak 的精确楔形市场 |
| Preply 英语学习细分 | 2025 | 全球 | USD 43.51B;年增长约 22% | 更广泛在线市场内的纯英语细分估算 | low | 方法论与 Technavio 不直接可比 |
| MarketsandMarkets 教育领域 AI | 2024-2030 | 全球 | USD 2.21B 至 USD 5.82B;CAGR 17.5% | 邻近教育 AI 市场规模估算 | medium | 范围过宽,不宜用作 Speak TAM |
| Speak 公司框架 | 2024 | 全球 | USD 100B+ | 公司主导的在线与线下语言学习框架 | low | 最宽泛且最不精确的视角 |
| TechCrunch 英语学习者 | 2024 | 全球 | 约 15 亿学习者 | 以 CEO 引言作为需求端人口视角 | medium | 学习者不等于付费用户 |
表格保留了各来源的范围差异,而非将其合并为单一混合 TAM。
[CM001, CM003, CM006, CM007, CM010, CM021]基于来源的 Speak 相邻市场规模参考点;低 / 中 / 高值重复出现,反映不同视角的点估算,而非概率情景。
各行在同一货币单位下比较不同口径的市场视角,不应加总或混合为 Speak 的单一 TAM。
[CM006, CM010]从宏观全球英语需求到 Speak 等口语优先付费产品窄域的概念性转化漏斗。
该漏斗混合人口与变现参考点,呈现收窄的商业现实,而非严格的统一预测;是辅助规模估算的概念工具,非收入模型。
[CM003, CM022]2.3 买家、用户与付费方细分
Speak 的买家图谱分为两条截然不同的销售路径。消费端路径中,用户即付费方——学习者按月支付订阅费进行 自主驱动的口语练习,TechCrunch 报道的月付 $20 或年付 $99 定价正契合这一自助模式。而在企业端路径中, 学员是用户,雇主是付费方,通常将英语培训作为提升效率、招聘吸引力、员工流动性或留用率的福利购买。 Speak 的 C 轮公告明确指出企业产品不限行业,且已声称拥有 200 家以上客户。 公开市场研究与这一拆分相吻合。Technavio 指出企业非学术学员是数字英语学习的主要细分群体,体现了英语 在国际商务沟通中的核心地位。Preply 的 2026 年报告同样指出,英语因商业和教育需求而成为默认的第二语言。 综合来看,消费端需求由情感和向往驱动,企业端需求由预算和生产力驱动。 两个细分市场的采购路径也不同。消费者可立即试用并订阅;企业则需要 HR、L&D 或部门负责人的预算审批, 并要求提供使用率、席位利用率和技能提升的证明。这使得企业端单客户经济价值可能更高,但销售周期更长, 续约依赖比消费端订阅更重。[CM009, CM023, CM024, CM025]
| 细分市场 | 买方 | 用户 | 付款方 | 工作流程/预算负责人 | 采用触发因素 |
|---|---|---|---|---|---|
| 消费者自助 | 个人学习者 | 个人学习者 | 个人学习者 | 应用商店订阅或直接应用内升级 | 旅行、职业流动、提升自信 |
| 雇主资助英语培训 | 学习与发展负责人/HR/业务部门经理 | 员工学习者 | 雇主 | 培训预算、生产力、留存或全球流动预算 | 跨境沟通需求或员工技能提升 |
| 学生/备考相关用户 | 学生或家长 | 学生学习者 | 学生/家长/学校 | 个人教育预算 | 为教育机会或证书目标学习英语 |
| 实时辅导替代品买方 | 个人学习者或雇主 | 学习者 | 学习者或雇主 | 辅导平台或合同预算 | 需要对话练习,但不打算建立内部培训项目 |
| 社区课程应用用户 | 个人学习者 | 个人学习者 | 个人学习者 | 通用学习应用订阅预算 | 比口语优先 AI 产品更便宜或更系统化的替代方案 |
公开数据在消费者和雇主用例方面较为充分,但针对 Speak 的正规学校采购数据匮乏。
[CM009, CM023, CM024, CM025]映射 Speak 各主要买家细分中谁审批支出、谁使用产品,以及需要哪些证明。
[CM009, CM023]2.4 增长驱动因素与采用制约
需求端的论据异常有力。英语是全球学习人数最多的语言,总使用者约 15 亿,在商业、教育和人员流动中具有 清晰的实用价值。Technavio 将市场增长部分归因于数字课程的灵活性,WEF 和 MarketsandMarkets 则着重强调 个性化学习、自适应内容及 AI 教育工具的更广泛转型趋势。Speak 的自有材料补充了一个更接地气的角度: 用户希望真正开口说话,而非单纯背记。对雇主而言,英语学习预算可作为效率提升或员工发展支出加以论证, 这强化了企业端的投资逻辑。 但公开来源也揭示了真实的制约因素。WEF 强调 AI 增强学习中存在的接入门槛、隐私、偏见和替代就业等问题。 世界银行认为,可扩展的 AI 采用依赖于连接性、算力、本地数据和技能——这些条件在新兴市场分布不均。 AI 语言工具的学术综述得出相似结论:个性化反馈有帮助,但隐私保护、证据质量和教师准备度仍滞后。 产品评测进一步锐化了风险点:Android Police 和 Languatalk 均认为 Speak 在激励用户方面表现出色,但 在严格的发音纠错和进阶水平反馈上偏弱。 实操层面的结论是:需求增长不等于无摩擦采用。Speak 受益于宏观需求,但其细分市场仍须证明学习效果、 可信的反馈质量和可接受的隐私规范,才能将兴趣转化为持久的付费使用。[CM004, CM008, CM012, CM015, CM016, CM017]
| 驱动因素 / 制约因素 | 方向 | 时间节点 | 含义 | 尽调问题 |
|---|---|---|---|---|
| 英语仍是商业与教育的默认语言 | 驱动因素 | 当前 | 支撑规模最大的单一语种学习需求池 | 按地区和语言对验证细分市场需求 |
| 移动端 / 自主学习灵活性 | 驱动因素 | 当前 | 助力自助订阅扩量,无需承担人工教师成本 | 按用户批次衡量学习频次和留存率 |
| AI 驱动的个性化与自适应反馈 | 驱动因素 | 当前 | 提升用户感知相关性,有助于提高参与度 | 索取有效性研究和完课率数据 |
| 雇主技能提升需求 | 驱动因素 | 当前 | 支持 B2B 销售运作,ACV 高于消费者订阅 | 索取企业席位数量和续约率 |
| 亚太地区增长集中 | 驱动因素 | 当前 | 有利于亚洲市场适配和分销能力强的企业 | 逐市场评估本地化能力和支付体系强度 |
| 数字鸿沟 / 基础设施缺口 | 制约因素 | 当前 | 制约低网络覆盖率和低收入市场的渗透 | 审查离线产品能力和国家市场构成 |
| 数据隐私、算法偏见与监管 | 制约因素 | 当前 | 可能延缓机构和雇主采用 AI 工具的进程 | 审查隐私架构及各司法管辖区专项政策 |
| 发音准确率不足或反馈浮浅 | 制约因素 | 当前 | 学习者过了入门新鲜期后,可能拉低留存率 | 委托第三方进行有效性与发音准确性基准测评 |
| 缺乏清晰的公开 SAM / 市场份额数据 | 制约因素 | 当前 | 估值叙事易构,承保叙事难立 | 索取内部市场份额及获客成本数据 |
尽调的核心分歧不在于需求是否存在,而在于变现能否做到高效且具有防御性。
[CM004, CM008, CM009, CM015, CM016, CM017]2.5 展示图表
03竞争格局
3.1 竞争格局概览:解决同一需求的多条路径
买家至少可以通过四类产品解决"帮我流利说一门外语"这一需求:宽泛的免费增值语言平台、专注 AI 口语的 应用、真人家教网络,以及结构化课程或社区驱动应用。Speak 属于口语优先的专注型集群,但也面临全部四类 产品的竞争,因为消费者和企业购买的不是产品品类,而是结果。这一点很关键:即便 Speak 的产品与众不同, 只要买家认为 Duolingo 够用、Cambly 更可信,或 Babbel/Busuu 更系统,Speak 就会失去份额。 公开规模层级一目了然。Duolingo 是品类头部,月活用户超 1 亿,具备平台级捆绑优势。Babbel 和 Busuu 的装机量远超 Speak,依靠系统化教学法而非 AI 新鲜感竞争。Cambly 占据真人家教替代赛道。ELSA、Praktika 和 Loora 证明 AI 口语赛道已趋于拥挤,而非蓝海。在这一格局下,Speak 的差异化主张不是"唯一选择", 而是"口语优先体验、AI 辅导与新兴企业分发的最佳组合"。 这是真实的定位,但尚未形成主导地位。[CP001, CP005, CP008, CP011, CP012, CP014]
| 竞争对手 | 类别 | 规模 / 融资信号 | 目标客群 | 差异化优势 | 局限性 |
|---|---|---|---|---|---|
| Duolingo | 现有综合平台 | 月活用户逾 1 亿;教育类应用营收第一 | 面向全球大众学习者 | 规模庞大、免费增值漏斗,支持 40 余种语言 | 并非纯口语优先;对话深度专注度偏低 |
| ELSA Speak | 直接专注口语竞品 | 下载量逾 1800 万;App Store 评分逾 10.9 万条 | 寻求发音和口语练习的英语学习者 | 以发音为核心的 AI 辅导 | 主要聚焦英语,非宽泛多语种平台 |
| Cambly | 真人在线家教替代品 | 全天候真人教师网络 | 重视母语者对话的学习者 | 真人问责机制与信任感 | 真人成本模式更贵且规模化难度较大 |
| Busuu | 社区驱动的系统化学习应用 | 用户逾 1.2 亿;Google Play 下载逾 5000 万 | 寻求系统化课程与社区反馈的学习者 | 母语者社区与系统化复习机制 | AI 对话专注度低于 Speak |
| Babbel | 系统化课程应用 | 订阅销量达 2500 万;Google Play 下载逾 5000 万 | 寻求专家设计课程的学习者 | 教学结构与复习循环 | 对话优先程度较低,以课程为主导 |
| Praktika | AI 教师新兴挑战者 | 学习者逾 2000 万;主打低价定位 | 追求 AI 教师式练习的价格敏感学习者 | 低价 AI 教师替代品 | 企业级分销渠道公开证据不足 |
| Loora | AI 英语教师挑战者 | 纯英语 AI 教师定位 | 职场人士与商务英语学习者 | 全天候英语对话教练 | 品牌覆盖面窄于大型现有玩家 |
来源未公开披露融资或收入数据时,表格以月活用户数、下载量或订阅销量等规模指标替代。
[CP005, CP008, CP011, CP012, CP014, CP016]| 购买标准 | Speak | Duolingo | ELSA | Cambly | Busuu | Babbel | Praktika / Loora |
|---|---|---|---|---|---|---|---|
| 口语优先工作流 | 强 | 中等 | 强 | 强 | 中等 | 中等 | 强 |
| 明确发音纠正 | 中等 / 存疑 | 中等 | 强 | 取决于教师 | 中等 | 中等 | 强 |
| 课程结构 | 中等 | 强 | 中等 | 低 | 强 | 强 | 中等 |
| 真人对话 | 否 | 否 | 否 | 是 | 仅限社区 | 否 | 否 |
| 多语言广度 | 中等 | 非常强 | 低 | 低 | 强 | 强 | 低至中等 |
| 企业 / 雇主市场拓展 | 有 | 公开证据有限 | 有但非核心 | 有 | 有限 | 有限 | 有一定商业推广,但验证不足 |
| 进阶学习者反馈深度 | 存疑 | 中等 | 中等至强 | 取决于教师 | 中等 | 强 | 中等 |
序数评级综合官方定位与独立评论。仅当公开证据具有方向性时,未知项才转化为保守的中级评估。
[CP001, CP008, CP011, CP023, CP033]Speak 主要竞争对手规模 / 分发能力与口语专精深度的序数定位图。
坐标轴为基于证据的序数评分,来源于公开规模指标、产品定位与评测专精信号;非调研数据。
[CP005, CP029]3.2 竞争对手画像与商业模式差异
Duolingo 是基准参照,原因在于它兼具庞大的免费用户规模、应用商店顶级营收排名和广泛的语言覆盖。它并非 Speak 口语优先辅导的最近产品类比,但在买家寻求"一款学语言的应用"时,是最强势的默认替代品。ELSA 才是更贴近的直接参照:同样以 AI 为核心、聚焦英语、以口语和发音为中心。其声称的 1,800 万次以上下载量 和 10.9 万条 App Store 评分,使其规模量级与 Speak 更为接近。 Cambly 依靠不同的劳动力模式竞争。它提供全天候真人母语家教,运营成本更高,但对于信任人类胜过 AI 的 学习者而言可信度更强。Busuu 和 Babbel 又有所不同:它们主打结构化体系、社区互动和课程深度,口语优先 程度不及 Speak,但教学体系更清晰。Praktika 和 Loora 展示了 AI 口语细分市场本身填充的速度——以更低 定价或更聚焦的英语家教定位抢占空间。 关键的规律不是某个竞争者在所有维度上都占优,而是 Speak 面对一组各有一技之长的替代品:规模、家教 可信度、结构化课程、社区属性,或低价 AI 辅导。[CP005, CP008, CP011, CP012, CP014, CP016]
呈现各竞争对手在规模、导师信任度、课程结构、社区与 AI 口语专精五个维度的相对强弱。
[CP011, CP032]3.3 能力、定价与分发对比
公开定价数据不完整,但方向性图像已足够说明问题。Speak 公开报道的消费端定价约为月付 $20 或年付 $99。 Praktika 将 AI 家教替代品定价约为月付 $8。ELSA 提供月付和年付会员。Duolingo 依托免费增值规模,付费 转化率约 9%,兼具广泛触达和定价灵活性。这一组合意味着双向的定价压力:低价 AI 辅导从下方挤压, 免费增值漏斗的巨大规模从上方压制。 能力层面,Speak 在买家重视流畅口语体验和低社交摩擦时最具优势。但在明确纠错方面,ELSA 的发音优先定位 与 Speak 自身的负面评价暗示 ELSA 可能更强。课程体系和复习机制方面,Babbel 和 Busuu 更扎实。真人督学 方面,Cambly 更强。纯分发规模方面,Duolingo 遥遥领先。 Speak 唯一明显差异化的商业角度是 Speak for Business。大多数公开竞争者仍主要是消费端产品。但由于 公开材料未披露客户名称、席位数或续约情况,该渠道的竞争重要性目前仍更多是潜力展望,而非已验证结论。[CP003, CP004, CP006, CP007, CP010, CP030]
| 公司 | 公开价格 / 模式 | 定位说明 | 未知项 | 含义 |
|---|---|---|---|---|
| Speak | $20/月或 $99/年(据报道) | 消费者订阅;口语优先 AI 教师 | 当前地区定价及企业平均售价未公开 | 中端消费类 AI 教师 |
| Duolingo | 免费增值加付费升级 | 庞大免费漏斗;付费版去除广告并增添功能 | Super/Max 当前具体定价因地区而异 | 对获客成本和支付意愿形成天花板压力 |
| ELSA | 月度和年度会员制 | 英语口语与发音辅导 | 当前实际平均售价不明 | 英语细分赛道直接订阅基准 |
| Cambly | 在线辅导订阅 | 真实母语者对话 | 具体套餐价格在已查阅来源中未获取 | 人工信任溢价替代品 |
| Busuu | 免费增值加订阅 | 系统化课程加社区 | 当前套餐价格未在此处获取 | 以更广泛的价值套装参与竞争 |
| Babbel | 完整课程需订阅 | 专家设计的系统化课程 | 当前结算价格因套餐而异 | 品牌强势的系统化课程替代品 |
| Praktika | 约 $8/月的营销主张 | 替代真人教师的 AI 教师 | 实际套餐结构和年度定价不明 | 对 Speak 形成低价 AI 竞争压力 |
本表比较公开标价锚点和套餐结构,而非实际净定价或企业合同价值。
[CP003, CP006, CP010, CP016, CP034]3.4 护城河持久性、切换成本与主要威胁
Speak 公开可见的护城河属于中等水准。公司拥有真实的产品身份——口语优先 AI 辅导——以及企业客户这一 潜在价值可观的第二渠道,这是正面因素。但护城河的持久性并不显著,因为每一项优势都面临可攻击的弱点: Duolingo 可以砸钱拉开分发差距;Cambly 可以凭人类可信度胜出;ELSA 可以在明确发音辅导上取胜;Busuu 和 Babbel 可以靠体系与留存循环拿分;Praktika 和 Loora 可以用更低价或更聚焦的定位压缩 AI 口语细分 赛道。 切换成本在公开证据中也看似有限。消费者能以低成本测试多款应用,且很可能确实如此。企业部署后可能面临 更高摩擦,但目前没有公开证据表明 Speak 拥有排他合同或深度工作流锁定。公开评测强化了这一风险:若 进阶用户感觉反馈浅薄或过于宽容,替代品唾手可得。 主要的近期威胁因此不是技术层面的颠覆性飞跃,而是慢慢积累的商业压力:定价挤压、分发捆绑和品类同质化。 Speak 仍有胜出的可能,但需要的是结果证明和企业端数据,而非更多营销新意。[CP020, CP022, CP024, CP025, CP026, CP027]
| 护城河主张 | 威胁 | 严重程度 | 缓解措施 / 尽调问题 |
|---|---|---|---|
| 口语优先 AI 用户体验 | 竞争对手能快速复制用户体验模式和课程流程 | 重大 | 以真实学习成效检验,而非依赖用户体验独特性 |
| 企业分销渠道 | 雇主多平台采购,或将 Speak 视为多种工具之一 | 高 | 索取客户背书、席位数量和独家条款 |
| 多语言拓展 | 大型现有玩家仍提供更多语言和更大生态 | 重大 | 追踪新语言上线节奏和各语言附加率 |
| 发音与反馈可信度 | 独立评测认为 Speak 反馈过于宽松 | 高 | 以 ELSA 和真人教师为基准测评口语准确性和反馈质量 |
| 品牌信任与账单体验 | 退款 / 支持投诉削弱支付意愿 | 中等 | 审计客服 SLA、退款率和投诉解决指标 |
| AI 细分市场防御性 | Praktika、Loora 及未来进入者压缩 AI 口语差异化空间 | 高 | 监控各 AI 教师竞品的定价动向和功能差异收窄情况 |
| 现有规模玩家压力 | Duolingo 免费漏斗和应用商店主导地位推高获客成本 | 极高 | 衡量获客成本(CAC)、回本周期及自然获客相对韧性 |
严重程度反映 Speak 差异化定位在未来两至三年内面临的商业风险。
[CP020, CP025, CP026, CP028, CP030, CP034]Speak 竞争持久性核心优势与风险的简明评分卡。
[CP020, CP041]3.5 展示图表
04财务情况
4.1 收入模式:表面混合,内部不透明
Speak 的公开财务状况在高层面清晰,底层却薄弱。公开证据支持两条变现路径:通过移动应用商店销售的消费端 订阅,和通过 Speak for Business 销售的企业端合同。TechCrunch 持续报道消费端定价约为月付 $20 或年付 $99, App Store 印证了月付和年付订阅的存在。与此同时,企业网站和 C 轮材料表明 Speak 已构建面向雇主的产品, 并声称拥有 200 家以上品牌或客户。 这足以判断 Speak 不只是一款消费端应用。但不足以量化收入结构、实际定价或收入质量。没有任何经审阅的 公开来源披露 ARR、付费用户数、转化率、退款率、流失率或企业收入占比。结果是一个常见的尽调陷阱: 变现架构可见,经济模型不可见。 行业背景也很重要。语言学习买家可以选择免费或语言交换产品,也可以购买 Speak 这样的中价位订阅,或按 课时付费给真人家教,或选择其他 AI 辅导工具。这意味着 Speak 的公开定价位于一条拥挤的价格走廊之中, 而非凌驾其上。[CI001, CI002, CI003, CI004, CI013, CI015]
| 收入来源 | 机制 | 计量单位 | 当前公开状态 | 质量 | 尽调问题 |
|---|---|---|---|---|---|
| 消费者订阅 | 通过应用商店销售月度 / 年度订阅 | 订阅用户 | 已确认 | 中等:标价公开,付费用户数未披露 | 订阅用户数、转化率、流失率、退款率及网页端与应用内账单占比 |
| Speak for Business 企业版 | 雇主 / 品牌合同 | 账户 / 席位 | 已确认 | 低至中等:买家数量有声明,经济指标未披露 | ACV、席位数、合同期限、续约率、实施成本 |
| 语言拓展上行空间 | 同一订阅延伸至更多课程目录,形成额外变现 | 语言附加率 / ARPU | 合理但尚未量化 | 低 | 各语言 ARPU、上线用户群体及按地区附加率 |
| 广告 / 赞助 | 广告变现 | n/a | 已查阅来源中未见踪迹 | 低 | 确认是否存在广告收入 |
| 市场 / 交易手续费 | 向教师或第三方收取抽成 | n/a | 已查阅来源中未见踪迹 | 低 | 确认是否存在抽成或合作伙伴收入 |
各行区分已确认的收入机制与合理但缺乏支撑的机制;缺失的收入来源明确标注,不推断填充。
[CI001, CI004, CI013, CI034]| 公司 / 模式 | 公开价格 / 单位 | 标价 vs 实际价格 | 未知项 | 来源 / 含义 |
|---|---|---|---|---|
| Speak 消费者端 | 约 $20/月或 $99/年 | 标价 | 促销折扣、地区定价、实际净定价均不明 | 中端 AI 订阅基准 |
| Speak 企业端 | 合同制企业定价 | 未公开 | 席位定价、合同期限、服务内容、折扣均不明 | 若企业业务占比显著,潜在收入质量更高 |
| Duolingo | 免费增值,付费升级 | 标价和定价模式已知,实际收入结构部分公开 | 当前各地区分层价格及利润率不详 | 为品类树立了强势低价锚点 |
| Cambly | 按课时计费,年付约 $8.12/课时 | 公开标价 | 教师结构、利用率及净收入不详 | 人工辅导可支撑更高单次消费 |
| italki | 市场化课时,试听价从 $5 起可见 | 按教师公开标价 | 抽佣比例、复购频率及完整课时均价不详 | 教师市场创造灵活消费阶梯 |
| HelloTalk | 免费语言交换 / VIP 生态 | 以免费入口为主打 | VIP 变现细节未见于所查资料 | 零价格替代品限制了轻度用户付费意愿 |
| Praktika | AI 辅导标价约 $8/月 | 营销定价 | 年付方案及实际均价不详 | 低价 AI 对 Speak 形成价格压力 |
本表比较各方公开价格锚点与变现结构,非实际 ARPU 或贡献利润率。
[CI002, CI019, CI020, CI021, CI025, CI038]| 指标 | 公开值 / 状态 | 置信度 | 重要性 | 尽调要求 |
|---|---|---|---|---|
| 消费者标价 | 约 $20/月或 $99/年 | 中 | 客户终身价值(LTV)的起点,但仅此不足 | 核实各地区当前定价及折扣路径 |
| 付费用户数 | 不可得 | 低 | 将下载量转化为收入的必要数据 | 要求提供当前及过去12个月付费用户数 |
| 毛利率 | 不可得 | 低 | AI 交付经济模型的核心验证指标 | 要求按渠道提供毛利率及算力 / 平台费用分项 |
| 获客成本(CAC)/ 回本周期 | 不可得 | 低 | 决定增长效率 | 要求提供获客渠道结构、CAC、回本周期及企业销售周期 |
| 退款率 / 客服负担 | 公开不可得;存在负面零星信号 | 低 | 消费者收入质量的重要指标 | 要求提供退款率、拒付率、客服积压及 SLA 指标 |
| 企业年合同价值(ACV)/ 席位经济 | 不可得 | 低 | 决定 B2B 业务是否改善整体模型 | 要求提供 ACV、席位数、合同期限、实施成本及续约率 |
| 留存率 / 净留存率(NRR) | 不可得 | 低 | 收入质量最重要的检验指标 | 要求提供队列留存率、总留存率及各细分市场 NRR |
公开数据如此稀疏正是关键所在:绝大多数决定性经济指标仍属私密。
[CI002, CI015, CI016, CI027, CI031]从用户获取与雇主获取到 Speak 两条可见收入轨道的定性桥接图。
该流程为定性桥接,公开来源仅确认收入轨道,未披露转化率、席位数或实际净收入。
[CI001, CI004, CI030]4.2 公开牵引力与定价参考:有用信号,弱核准依据
公开牵引力是真实的。Speak 宣传 1,500 万次以上下载量,App Store 显示庞大的评分基数,Google Play 显示六位数的评论量。TechCrunch 还报道日均使用时长为 10 至 20 分钟。这些都是有意义的信号,表明产品 分发广泛、活跃使用持续。 但公开牵引力与收入不是一回事。下载量无法告知活跃用户数、转化率或留存时长。评论数无法反映退款率或 净计费额。日均使用时长也无法区分免费试用用户、付费订阅者和企业赞助席位的贡献比例。同样的谨慎适用于 竞品定价页面:竞品标价有助于三角测算品类支付意愿,但不能反映实际平均销售价格(ASP)或贡献利润率。 不过,定价走廊仍有信息量。相比低价 AI 辅导,Speak 定位中端;明显低于反复真人辅导所需的支出;高于 免费或免费增值替代品。这支撑了变现合理性,同时也凸显了定价权的局限。[CI005, CI006, CI015, CI017, CI018, CI019]
4.3 成本结构、单位经济模型与资本充足性
从公开证据看,Speak 在生产端接近轻资产,但运营层面未必廉价。没有可见的存货、硬件或项目融资负担,公司 更像一个软件与内容业务。然而,AI 辅导引入了静态课程应用所不必完全承担的成本层:推理、语音处理、 反馈生成、本地化、客户支持和持续内容扩充。 这些成本驱动因素可能比资本支出更重要。应用商店还在分发和收款两端引入了平台依赖。企业端方面,任何 有实质意义的企业运营都会叠加销售、入职和客户成功成本。这些并非天然的问题所在,但在公开来源中均未 被直接量化。获客成本(CAC)、回本周期、毛利率和留存情况均属私有信息。 资本充足性因此是情景假设问题。公司显然近期已完成有意义的融资,且未出现公开的财务困境信号。但由于 当前现金储备和烧钱速度未披露,公开记录无法区分"资金充裕"和"已融资但仍依赖下一轮才能支撑招聘或 企业端扩张提速"。[CI008, CI010, CI011, CI012, CI027, CI028]
| 项目 | 公开状态 | 当前判断 | 重要性 | 下轮融资 / 尽调触发条件 |
|---|---|---|---|---|
| 近期融资 | 已确认 | 近期 B-3 轮及 C 轮累计公告融资约 $98M | 提供近期资金支撑 | 核对公告融资轮次与股权结构表及净募集金额 |
| Form D 佐证 | 部分确认 | Form D 追踪显示与2024年末融资相关的 $77.7M 发行 | 提高对公告融资实质完成的置信度 | 要求提供正式完成文件及融资时间表 |
| 手头现金 | 不可得 | Unknown | 最直接的现金跑道输入 | 要求提供当前余额及受限现金 |
| 月度烧钱速度 | 不可得 | Unknown | 决定现金跑道消耗速度 | 要求提供月度烧钱速度及各职能预算 |
| 现金跑道月数 | 不可得 | 仅为情景估算 | 公开资料无法支撑判断 | 基于实际烧钱数据构建基准 / 下行 / 扩张情景 |
| 债务 / 项目融资 | 未见公开证据 | 大概率没有,但未经充分核实 | 债务会改变风险结构与财务灵活性 | 要求提供债务时间表、租赁及或有义务 |
本章重点评估资本充足性,不涉及已在其他章节覆盖的历史融资轮次详情。
[CI008, CI009, CI010, CI012, CI033, CI035]| 缺失指标 | 对尽调的影响 | 具体尽调路径 |
|---|---|---|
| 营收 / ARR / 账单收入 | 无法评估规模或增长质量 | 要求提供经审计或董事会级别的过去12个月收入桥接数据 |
| 订阅用户数及转化率 | 无法将下载量与收入挂钩 | 要求提供订阅漏斗、试用转化率及套餐结构 |
| 毛利率及平台费用结构 | 无法评估 AI 交付经济模型 | 要求按算力、内容、客服及应用商店费用分项提供营业成本 |
| 企业 ACV / 席位数 / 续约 | 无法评估 B2B 收入流质量 | 要求提供头部客户合同摘要及续约记录 |
| 烧钱速度与现金跑道 | 无法评估融资依赖度 | 要求提供月度现金余额及烧钱预测情景 |
| 退款 / 流失 / 客服质量 | 无法评估消费者收入漏损 | 要求提供退款率、各套餐流失率及投诉解决指标 |
具体尽调路径已纳入交付物,数据缺口是可操作的问题,而非修辞层面的表述。
[CI007, CI014, CI027, CI036]即便数值缺失,也展示经济上重要的公开输入项。
多数节点是公开概念而非公开数字;桥接图旨在使缺失指标显性化。
[CI015, CI027, CI029]序数区间视图,展示 Speak 公开财务叙述中哪些部分相对可见、哪些几乎空白。
这些为基于证据的序数评分,非管理层 KPI。评分概括了公开披露的程度:定价与融资相对可见,营收与现金跑道仍基本未披露。
[CI002, CI008, CI007, CI012, CI036]基于公开证据,映射主要现金驱动因素。
该矩阵为定性表述,由产品与分发模型佐证,非披露成本科目。
[CI028, CI033, CI037]4.4 财务结论:变现架构有潜力,收入证据有限
财务结论是:对架构谨慎乐观,对透明度谨慎偏负面。Speak 显然拥有可变现的消费端和企业端界面、近期的 机构融资,以及应当能支撑一定经常性收入的产品使用数据。公司所在品类中,消费者已习惯为辅导、订阅和 备考付费,需求是否存在不是问题。 问题在于公开记录验证的是架构,而非表现。估值从 5 亿美元升至 10 亿美元,背后是快速融资势头的推动, 但财务披露并未同步跟上。公开证据无法回答:业务实际产生多少收入、应用商店费用和 AI 交付成本之后的 毛利率是多少、订阅者的粘性如何,以及企业账户是否实质性地改善了经济模型。这使得核准依据过度依赖 融资头条、用户代理指标和市场情绪。 尽调的下一步不是找更多营销页面,而是获取内部财务资料包:ARR 与计费额、付费用户数、各渠道毛利率、 退款与流失数据、企业年度合同价值(ACV)、销售效率,以及真实的现金跑道模型。[CI032, CI034, CI035, CI036]
4.5 展示图表
05产品与技术
5.1 产品界面与学习者工作流
Speak 的公开产品叙事异常连贯:该应用被定位为口语优先的 AI 家教,而非词汇游戏或语法参考工具。无论是 官网首页、应用商店描述还是产品更新帖,同一套工作流反复出现:学习者进入结构化课程,大声朗读目标句, 立即获得纠正反馈,再将练习内容应用于更自由的 AI 对话中。这套工作流现已被正式化为 Speak Method 的 "学习→练习→应用"循环。在实际操作中,可见的模块包括家教课程、口语练习/口语卡片、角色扮演、自由 对话、学习进度追踪,以及单元复习和 Speak Level 等较新的复习与水平评估功能。产品界面仍以移动端为主, 但已不再只是英语学习应用——公开帮助内容显示面向学习者的六种目标语言和 15 个英语教学的母语入口。 核心战略结论是:Speak 在教学法与产品包装之间实现了紧密的工作流契合,每个主要模块最终都服务于同一 目标——让学习者开口、对比输出、不断重试,而非被动消费内容。[CE001, CE002, CE004, CE005, CE006, CE008]
| 模块 / 资产 | 主要用户 | 状态 / 成熟度 | 差异化特点 | 尽调缺口 |
|---|---|---|---|---|
| Tutor Lessons | 新学员及进阶学员 | 核心 / 已上线 | 动态 AI 辅导员能纠错、答疑、引导学习方向,而非仅播放预录音频 | 无按模块划分的完课率或学习成果公开基准 |
| 口语练习 / 口语卡片 | 所有参与结构化课程的学员 | 核心 / 已上线 | 以口语为先的重复练习,搭配内部研发的匹配算法与音素对齐,不同于通用题库机制 | 匹配准确率及误报控制的公开证据仅来自内部 |
| Live Roleplays | 进入开放式练习阶段的学员 | 已上线,但2024年末最初为有限推出 | 实时语音对话将 OpenAI 语音直通语音能力与 Speak 熟练度图谱、提示及学习目标相结合 | 按市场 / 套餐划分的公开推出范围未量化 |
| Free Talk / 自定义对话 | 自主学习者及高意向练习用户 | 已上线 | 开放式 AI 对话个性化练习场景,允许用户围绕自身情境创建练习内容 | 第三方评测指出对话控制力及反馈深度仍难满足深度学习者需求 |
| 进度 / 复习功能(Unit Refreshers、Speak Level、支线任务) | 课间回访的活跃学员 | 已上线 / 按市场分阶段推出 | 在口语工作流基础上增加留存循环与可量化的流利度追踪 | 进入2026年,Speak Level 仍仅向部分市场的英语学习者开放 |
公开信息呈现出完整的模块体系,但多项较高层次的质量主张仍属内部数据或推出范围有限。
[CE004, CE005, CE006, CE008, CE016, CE027]| 用户任务 | 当前工作流 / 痛点 | Speak 解决方案 | 可观察的公开效益 | 局限 / 注意事项 |
|---|---|---|---|---|
| 快速开口说话 | 传统应用过度侧重阅读、语法或词汇识别 | 以口语为先的入门流程,即时重复练习与反馈 | 官网及评测多次提及学员从第一天起即可开口说话 | 效果多为用户自述,缺乏第三方疗效测试支撑 |
| 练习真实对话 | 人工辅导价格高且难以预约 | Roleplays 与 Free Talk 借助 AI 搭档模拟常见场景 | 用户和第三方评测均强调真实对话练习的价值 | 部分评测指出 AI 有时显得重复或过度追问 |
| 纠正发音或表达 | 通用 ASR 难以捕捉带口音的学员语音 | 自研 ASR 搭配匹配技术栈,在音频细节敏感场景中引入语音直通语音 | Speak 报告后端升级后反馈更快、WER 更低 | 多数性能数据属于内部指标,未经外部基准测试 |
| 中断后续学 / 保持学习动力 | 语言学习应用经常在两次课程之间流失用户 | Unit Refreshers、连续学习系统、支线任务及 Speak Level 进度展示 | 2025年冬季版本明确新增留存与进度功能入口 | 部分应用商店评测仍要求更丰富的奖励机制与游戏化循环 |
| 英语母语者学习其他语言 | 早期 Speak 产品聚焦亚洲英语学习者 | 2025年语言扩展在西班牙语之后新增法语、日语、韩语和意大利语 | 公开帮助页面现已收录六种目标语言 | 旗舰课程以外的中级深度内容仍不完整 |
口语工作流在 Speak 能将学员锁定于「说—听—纠错」紧密闭环的场景中最为有效;开放式掌握与内容广度仍有待充分验证。
[CE001, CE004, CE008, CE009, CE015, CE018]该栈聚焦于面向学习者的模块、教学逻辑、语音反馈系统与支持运营如何叠加,构成 Speak 的口语优先用户体验。
[CE001, CE004, CE008, CE016, CE029, CE033]5.2 技术架构与运营模式
Speak 现在披露的技术细节已超过大多数消费端教育应用。2024 年的 ASR 系统重构,将公司从分散的设备端和 第三方语音系统,整合至统一的后端技术栈。Speak 称该技术栈在学习者语音数据上微调了 Conformer-CTC 模型, 以 Nvidia Riva/Triton 在 Kubernetes 和 Google Cloud 上提供推理服务,并借助 gRPC 加 WebSocket 流式 传输,将语音反馈的延迟压低至课程互动所需的水平。2025 年,Speak 推出 Matching v2,将流式 ASR 与音素 模型和强制对齐结合——这一点至关重要,因为学习者语音往往会打破通用 ASR 的内置假设。到 2026 年初, Speak 在此基础上叠加了更宽泛的语音智能体平台:移动应用通过 LiveKit Cloud 以 WebRTC 方式连接; 语音智能体服务器编排外部 ASR、LLM、TTS 和语音转语音服务商;系统按功能特性选择级联模式或端到端语音 转语音模式,而非强制使用统一技术栈。这一架构体现了真实的技术深度,但也带来了真实的运营复杂度。 Speak 明显是多服务商、区域感知、以延迟为重要指标的系统,公开讨论涵盖故障转移、尾延迟监控、语义轮次 检测和在质量或可用性下降时切换服务商等细节。[CE010, CE011, CE012, CE013, CE014, CE015]
| 层级 / 流程 / 组件 | 职责 | 关键依赖 | 主要风险 |
|---|---|---|---|
| iOS / Android 移动客户端 | 采集学员音频、播放回应、承载课程体验 | 原生移动应用;操作系统权限;当前版本 | 无桌面端兜底,麦克风权限问题可导致语音流程中断 |
| 实时传输层 | 客户端与后端间的低延迟双向音频 | 通过 LiveKit Cloud 的 WebRTC;区域路由 | 外部平台依赖及跨区域延迟波动 |
| 语音智能体运行时 | 执行课程逻辑、轮次管理、提示、纠错及编排 | LiveKit Agents 加 Speak 应用逻辑 | 功能数量与服务商排列组合增加,复杂度随之上升 |
| 语音识别与匹配 | 转录学员语音并映射至课程目标 | 微调 Conformer-CTC、Nvidia Riva/Triton、wav2vec2 音素模型、强制对齐 | 内部指标表现强劲,但未经外部审计;带口音的边缘场景仍存在 |
| LLM / TTS / 语音直通语音服务商 | 生成辅导员回应与语音输出 | OpenAI Realtime API、外部 LLM/TTS 服务商、备用服务商路由 | 多服务商成本、可靠性与模型质量差异 |
| 学习引擎与内容系统 | 存储课程、熟练度图谱、对话状态及分析数据 | Speak 后端服务与课程系统 | 数据治理、评估及版本管理规范的公开证据有限 |
| 可观测性与故障切换 | 监测延迟 / 错误,并在服务商降级时重新路由流量 | 各服务商指标、备用路由、区域感知运营 | 公开披露未量化正常运行时间或历史故障情况 |
Speak 的技术披露表明其构建了真实的生产级 ML 技术栈,而非对单一模型供应商的薄层封装。
[CE011, CE012, CE016, CE017, CE019, CE020]| 日期 / 阶段 | 功能或里程碑 | 状态 | 含义 | 来源 |
|---|---|---|---|---|
| 2024-06 | 后端 ASR 全面升级 | 已发布 | 核心语音反馈迁移至更强的统一后端,并进行学员语音内部调优 | Speak ASR 博客 |
| 2024-10 | 借助 Realtime API 的实时角色扮演 | 已发布,初期有限推出 | 推动 Speak 更接近自然实时对话,但对 OpenAI 的依赖也随之加深 | Speak Live Roleplays 博客 |
| 2024-11 | Google Play 在香港 / 韩国 / 台湾获得最佳应用认可 | 已发布 / 外部认可 | 表明 Speak 起步的核心亚洲市场获得了用户认可 | Speak Google Play 博客 |
| 2025-06 | 面向英语母语者的法语、日语、韩语和意大利语上线 | 已发布 | 将总可用市场(TAM)扩展至英语学习场景之外 | Speak 新语言博客 |
| 2025-12 | 冬季版本:自适应课程、复习练习、支线任务、音频角色扮演、Speak Level | 已发布 / 部分分阶段推出 | 新增留存与熟练度功能入口,而非仅限新增课程 | Speak 冬季发布博客 |
| 2026-03 | Voice Agent Platform 披露 | 已发布平台能力,仍在持续演进 | 表明在基础设施、轮次检测及更丰富语音直通语音应用方面持续投入 | Speak 语音智能体平台博客 |
| 2026年持续进行 | 更多中级课程、更多语言、更广泛的 Speak Level 推出 | 开发中 | 广度与成熟度仍是动态目标,而非完整产品 | Speak 帮助中心和发布说明 |
Speak 发版频繁,但多个高价值路线图项目仍明确处于开发中或受市场限制。
[CE008, CE009, CE015, CE019, CE027, CE049]5.3 可信度、质量与支持控制机制
在信任与运营质量方面,Speak 的公开证据参差不齐,但整体优于纯营销导向的应用。公司暴露了具体的支持 流程:应用内问题反馈、故障排查指南、麦克风权限核查清单,以及未解决问题的邮件升级通道。应用商店公开 内容还披露了基本的隐私和平台控制,包括数据传输加密、删除请求、最低操作系统支持和仅限移动端可用。 课程质量管控也比一般 AI 应用更明确:Speak 表示课程由学习设计师编写,AI 用于加速生产,最终内容仍由 人工审核。信任图景中较弱的一面是外部摩擦。独立评测称赞了语音质量和以口语为中心的设计,但也报告了 反馈深度浅薄、高级套餐分层混乱、语音识别间歇性卡顿、退款投诉和支持响应迟缓等问题。对尽调而言更 重要的是:Speak 的公开隐私政策和服务条款 URL 在纯文本审阅中无法呈现可读内容,我们也未能找到公开的 SOC 2、ISO 27001 或类似安全保障证明材料。这意味着尽管消费端支持机制清晰可见,企业端信任层面仍 存在真实缺口。[CE029, CE030, CE031, CE032, CE033, CE034]
| 控制措施 / 披露 / 质量信号 | 状态 | 范围 | 缺口或含义 |
|---|---|---|---|
| 课程由学习设计师编写并经人工审核 | 存在 | 适用于课程及课程创建 | 内容完整性的良好质量信号,但未发现外部教学法审计 |
| 应用内问题反馈标记加客服邮件工作流 | 存在 | 课程问题、账户问题、故障排查升级 | 表明运营支持流程存在,但以邮件为主,可能难以均匀扩展 |
| 语音识别故障排查指南 | 存在 | 麦克风权限、清除缓存、操作系统更新、Samsung/Bixby 冲突 | 证实公司预期语音采集故障会在实际使用中发生 |
| Google Play 上的数据处理披露 | 存在 | 声明共享应用活动 / 设备 ID、收集个人及财务信息、传输加密、删除请求 | 有用的基础披露,但不能替代详细的隐私或安全计划 |
| 隐私与条款网页 | 部分可查 | 应用商店列表中有公开 URL | 纯文本审查返回 JS 空壳,降低了直接尽调可见度 |
| 安全认证 / 审计 / 信任中心 | 公开未发现 | 章节证据中未收录公开的 SOC 2、ISO 27001、渗透测试摘要或正常运行时间页面 | 若承保企业扩展或受监管买家,属于重大尽调缺口 |
信任证据足以支撑消费者应用审查,但对深度企业或隐私尽调而言仍不完整。
[CE029, CE033, CE034, CE035, CE036, CE043]最重要的公开依赖不是每个内部组件,而是可能瓶颈性能、隐私审查或规模扩张的外部平台与接口。
[CE019, CE027, CE031, CE034, CE036, CE047]Speak 在核心口语 UX 与底层语音基础设施上最强;公开证据在正式信任材料与高阶学习者深度方面较弱。
[CE018, CE033, CE034, CE036, CE042, CE043]5.4 差异化、成熟度与待解尽调问题
Speak 最强的产品层面差异化不在于"AI 家教"的品牌标签,而在于一套专为语言学习者定制的语音技术栈, 与同类通用聊天机器人拼凑出的产品有本质区别。公开声称的学习者口音 ASR、音素匹配、多服务商 TTS 选择 和特性级语音管道设计均支持这一判断。OpenAI、TechCrunch、应用商店和独立评测者的外部印证表明,市场 看到了同样的规律:一款打磨精良的口语产品,用户爱用,技术野心真实。成熟度信号可信,但不应过度解读。 评分和下载量强劲,应用版本更新频繁,Google Play 的认可暗示在亚洲有真实的分发渗透。但若干核准问题 仍悬而未决:公司安全合规体系的深度、更丰富的语音技术栈在规模下是否经济高效、路线图中仍受制于推广 节奏的比例,以及独立学习效果证据是否存在于内部产品指标和客户证言之外。简言之,Speak 在产品和技术 差异化上表现出色,但公开证据在能力和体验层面远强于正式信任/合规披露层面。[CE023, CE024, CE037, CE038, CE039, CE040]
该流程聚焦 Speak 的教学循环,而非模块清单:引入语言、大声操练,再在真实对话中应用并回顾。
[CE005, CE031, CE039, CE045]5.5 展示图表
06客户情况
6.1 客户基础与细分
公开证据表明,Speak 目前首先仍是一款自助式消费端语言应用,企业雇主付费的覆盖层正在浮现,而非纯粹 的企业业务。官网首页和应用商店描述强调直接面向个人学习、六条核心语言学习路径、AI 口语练习和应用 商店规模。自助服务层显然已具规模:官方与第三方数据均显示,截至 2024 年年中学习者已逾 1,000 万, 截至 2025 至 2026 年底下载量约达 1,500 万。地域分布也很重要。韩国是 Speak 的首发市场,公司公开 表示 2024 年韩国约 6% 的人口在使用该应用学习英语,TechCrunch 早在 2023 年就报道了超过 10 万名 韩国付费用户。与此同时,Speak 已扩展至首发市场之外,官方与第三方报道均显示覆盖 40 多个国家, 语言产品涵盖英语、西班牙语、法语、意大利语、日语和韩语。新兴的 B2B 路径以雇主为付费方、员工为 主要用户,英语技能提升为旗舰使用场景。[CU001, CU002, CU003, CU004, CU005, CU006]
| 细分市场 | 买家 / 用户 / 付款方 | 使用场景 | 规模证据 | 收入 / 战略价值 | 缺口 |
|---|---|---|---|---|---|
| 自助消费者学员 | 个人 / 学员 / 个人 | 面向旅行、工作、学业及日常对话的口语练习 | Speak 官网显示超1500万次下载;44K iOS 评分;112K Google Play 评测 | 核心消费者订阅引擎,同时是品牌扩散的顶部漏斗 | 无公开的免费转付费转化率或付费用户构成 |
| 韩国英语学习群体 | 个人或雇主 / 学员 / 个人或雇主 | 在 Speak 起家市场提升英语流利度与自信 | 2024 年约占韩国人口 6%;2023 年韩国订阅用户逾 10 万 | 解释了早期用户密度、评价量和企业端初始需求的来源 | 韩国当前的收入与用户占比未对外披露 |
| 国际多语言消费者 | 个人 / 学员 / 个人 | 西班牙语、法语、英语、意大利语、日语和韩语口语练习 | 截至 2024 年覆盖 40 余国,主页提供六条核心课程线 | 推动韩国以外的全球增长,拓宽交叉销售空间 | 无分国家的用户或收入拆分 |
| 雇主付费 B2B 团队 | 雇主或学习与发展部门 / 员工 / 雇主 | 商务英语与员工语言能力提升 | B2B 页面列有 200 余个品牌;C 轮公告称企业客户逾 200 家、员工采用率达 85% | 叠加在消费者需求之上的更高 ARPU 扩张路径 | 无公开的 ACV、席位数或续约数据 |
| 有名有姓的企业订阅客群 | 雇主 / 员工 / 雇主 | 员工语言福利或技能提升项目 | 据 Forbes 报道,约 500 家企业(包括 KPMG 和 HD Hyundai)提供订阅,主要在韩国 | 表明已有实质性采购渗透,不只是试点 | 多数客户无公开名单、案例研究或部署结果 |
公开证据显示,Speak 以消费者为主要基础,雇主付费层级虽已形成规模,但仍未得到充分披露。
[CU002, CU004, CU005, CU006, CU009, CU012]| 指标 | 数值 | 日期 | 来源 | 置信度 | 含义 | 缺失分母 |
|---|---|---|---|---|---|---|
| 学习者 / 用户 | 10M+ | 2024-06-18 至 2024-06-20 | Speak Series B + TechCrunch | 中 | 2024 年中期,Speak 已实现有意义的全球规模 | 未区分免费、付费或活跃用户 |
| 下载量 | 15M+ | 2026-05-05 访问 | Speak homepage | 中 | 当前顶部漏斗规模已明显大于 2024 年的用户披露数字 | 无安装到激活的转化分母 |
| App Store 评分量 | 44K ratings | 2026-05-05 访问 | Apple App Store | 高 | 大量 iOS 评价证实消费者持续活跃 | 无分国家或付费类型的评分分布 |
| Google Play 评价量 | 112K 条评价 / 10M+ 下载 | 2026-05-05 访问 | Google Play | 高 | Android 端大量公开验证了产品的采用规模 | 无 DAU/MAU 或付费用户分母 |
| 口语台词数 | 3.74B | 2025-12-19 | Speak Wrapped 2025 | 中 | 反映用户持续进行口语练习,而非一次性安装 | 无唯一用户分母 |
| 应用内练习时长 | 19.6M hours | 2025-12-19 | Speak Wrapped 2025 | 中 | 反映大量学习时长累积与习惯养成 | 无分付费用户或队列的人均时长 |
| 已开始课程数 | 231M | 2025-12-19 | Speak Wrapped 2025 | 中 | 表明规模化的课程重复消费 | 无完成率分母 |
| 已创建个性化课程数 | 80.3M | 2025-12-19 | Speak Wrapped 2025 | 中 | 表明用户积极使用自适应或个性化路径 | 无参与个性化功能的用户占比 |
| 企业客户数 | 200+ | 2024-12-10 | Speak Series C + Dataconomy | 中 | 证实 B2B 已超出落地页试验阶段 | 无分细分市场、地区或合同规模的拆分 |
| 提供员工订阅的企业数 | ~500 | 2025-11-12 | Forbes | 低 | 表明企业分销广度超出官方案例研究数量的暗示 | 无活跃席位或付费客户分母 |
上述指标均为采用与活跃度代理,而非留存队列或收入质量指标。
[CU004, CU009, CU010, CU015, CU017, CU018]Speak 的公开用户旅程始于个人发现应用、完成首次口语练习,延伸至习惯养成、应用商店口碑积累,部分用户最终推动雇主为其买单。
[CU001, CU004, CU007, CU026, CU028]公开证据支持一个可复现的流程:从消费者发现应用到反复口语练习;其中有一条更窄的分支通向雇主付费部署。
[CU004, CU010, CU013, CU017, CU018, CU019]6.2 具名证明与采用信号
最强的公开证明不是一系列企业案例研究,而是大规模评测数据与有名有姓的学习者证言的组合。Apple 和 Google 均显示大量评分和评论数量,Speak 自有的评测页面复用了在公开应用商店界面上同样出现过的可识别 客户评论。这种重复具有价值,因为它降低了所有引言完全合成的风险,尽管官方页面仍是经过正向筛选的 内容。最清晰的具名示例来自消费端学习者。j herronov 表示数月的法语学习提升了对法语媒体的理解力, 让自由对话和书签功能变得有价值;Dan S 表示超过六个月的西班牙语学习优于此前工具,因为反馈详细且 即时;Rosalyn Mulder 表示 Speak 的家教式反馈和打卡连胜功能使其成为比 Duolingo 和 Mango 更优的 选择。这些仍是自述结果,但具体、近期,且在独立分发渠道中均可见。企业端证明较薄但并非不存在: Speak 称拥有 200 家以上企业客户,Forbes 报道约 500 家公司(包括 KPMG 和 HD Hyundai)向员工提供 订阅,主要集中在韩国。[CU005, CU008, CU015, CU016, CU023, CU024]
| 客户 | 细分 | 部署 / 使用场景 | 生产 vs 试点 | 结果 | 局限性 |
|---|---|---|---|---|---|
| j herronov | 消费者法语学习者 | 使用 Speak 数月,提升听力、自由对话能力,并收藏片段用于真实对话 | 消费者主动使用 | 称该应用改善了在法语媒体中捕捉词汇和表达的能力,并提供了实用的个性化练习 | 自我报告结果;无独立能力评估 |
| Dan S | 消费者西班牙语学习者 | 使用 Speak 超过 6 个月作为主要口语反馈工具 | 消费者主动使用 | 称详细且即时的反馈使 Speak 优于此前的西班牙语学习选项 | 单一评测者,且无量化的前后基准测试 |
| Rosalyn Mulder | 消费者西班牙语学习者 | Android 用户,将 Speak 与 Duolingo 和 Mango 对比 | 消费者主动使用 | 称导师式反馈和连续学习机制提升了积极性,使 Speak 成为首选 | 仅有动机层面的证明;无留存或消费披露 |
仅为样本。消费者评测者中具名的公开证明远比企业客户丰富。
[CU031, CU032, CU033, CU048]Speak 消费端公开口碑扎实,企业客户数量也有公开披露,但留存和客户集中度几乎没有公开数据。
[CU008, CU016, CU029, CU043, CU044, CU045]6.3 持久性、扩张与集中度风险
关键的尽调问题在于:客户质量证据远弱于采用规模证据。公开材料展示了下载量、评分、课程数、已开口 句数,甚至企业使用率,但未披露净留存率(NRR)、总留存率(GRR)、流失率、合同期限、队列留存或 头部客户集中度。这意味着 Speak 能可信地展示谁在试用和喜爱产品,但无法证明这些客户在收入层面的 持久性。变现界面还引入了流失分析中至关重要的摩擦点:应用商店定价显示自助订阅从月付约 $18 到年付 约 $84 的高级计划和更高端的高级加量计划不等,而独立投诉页面则反映了退款、自动续费、语言覆盖和 语音识别方面的投诉。这些并不能证明系统性弱点,但确实说明公开记录在展示获客和使用数据上更胜一筹, 续约质量则付之阙如。企业扩张故事也颇有说服力但尚不完整:Forbes 称这一布局起步于消费者要求雇主 买单,这是健康的由用及企业信号,但公开语料库仍缺乏具名案例研究库、合同条款,或能够判断 B2B 收入是否仍高度集中于韩国的地理拆分。公开证据也未能将头部活动指标与活跃付费用户数打通,投资者 仍需要基础数据:付费用户队列、年付计划续约曲线、退款率,以及按账户/地域分层的企业留存。在这些 数据私下披露之前,正确的解读是:Speak 已展现需求与使用习惯,但尚未能全面审计各队列、各客户和 各地域的客户持久性。[CU016, CU026, CU028, CU036, CU037, CU039]
| 指标 | 数值 | 细分 | 置信度 | 尽调要求 |
|---|---|---|---|---|
| App Store 满意度 | 4.8 | iOS 消费者用户 | 高 | 请求分国家、语言课程和付费状态的评分分布 |
| Google Play 满意度 | 4.7 | Android 消费者用户 | 高 | 请求评分分布及卸载 / 退款关联数据 |
| 员工采用率 | 85 | Speak for Business 员工 | 中 | 请求指标背后的队列定义、席位基数和时间窗口 |
| 增长耐久性代理 | 用户五年间逐年翻倍,至 2024 年 | 全部用户 | 中 | 请求付费用户留存数据,以区分增长与流失掩盖效应 |
| 重复使用代理 | 2025 年口语台词达 3.74B 条 | 全部用户 | 中 | 请求活跃用户分母和队列留存曲线 |
| 公开 NRR | 企业 | 低 | 请求分合同队列和地区的净留存率(NRR) | |
| 公开 GRR / 流失率 | 全部细分 | 低 | 请求客户流失、席位流失和非自愿计费流失数据 | |
| 公开合同期限 / 续约情况 | Speak for Business 企业版 | 低 | 请求中位合同期限、续约率和分客户的扩张率 |
本表替代原计划的队列图,原因是未有公开来源提供适合绘制队列图的分时间段留存百分比。
[CU016, CU023, CU024, CU040, CU045, CU046]| 扩张驱动力 | 集中度 / 摩擦风险 | 影响 | 尽调路径 |
|---|---|---|---|
| 消费者向雇主增购 | B2B 采用可能仍主要由韩国消费者需求播种 | 若韩国仍是主要验证市场,可能高估全球企业端的可复制性 | 请求分国家的企业 ARR、席位数和续约情况 |
| 口语优先差异化与高评分 | 公开评分无法反映付费队列是否以有吸引力的比率续约 | 采用质量可能优于收入耐久性 | 请求年度计划续约曲线和退款率 |
| 更多国家和语言 | 语言目录投诉显示部分用户希望支持更多语言或改善本地化 | 可能限制相邻学习者细分市场的 TAM,并在初次试用后提高流失风险 | 请求语言路线图、候补名单规模及分新语言的留存数据 |
| 雇主付费采用指标 | 85% 的员工采用率有潜力,但缺乏席位数和合同价值的背景信息 | 可能反映的是规模较小但高度活跃的子集,而非规模化企业扩张 | 请求分客户的席位激活率和 ACV 分布 |
| 自助订阅定价 | 自动续订和退款投诉可带来客服和非自愿流失负担 | 消费者流失和品牌口碑可能比评分所显示的更为波动 | 请求分应用商店的退款、取消和拒付指标 |
| 企业客户总数 | 200+/500 家企业客户无公开名单,集中度无法核查 | 少数大客户可能贡献了不成比例的 B2B 收入份额 | 请求前 10 大客户的收入占比和赞助商集中度 |
公开记录支持扩张叙事,但无法提供清晰的集中度全貌。
[CU026, CU028, CU029, CU037, CU041, CU042]| 来源 | 信号类型 | 内容 | 倾向 | 局限性 |
|---|---|---|---|---|
| Apple App Store 评价 | 客户证明 | 大量近期评价流,评分高且有具体功能反馈 | 可访问样本中以正面为主 | 无完整评分分布或退款数据 |
| Google Play 评价 | 客户证明 | 大量 Android 评价,并可见识别准确性和语言处理的投诉 | 正面与负面案例均可见 | 无付费状态或留存关联 |
| Speak 官方评价页 | 精选客户证明 | 复现真实具名评价,便于检索 | 正向筛选,仅限 App Store 五星评价 | 不能单独作为平衡的满意度读数 |
| LanguaTalk review | 独立评测 | 对反馈深度、课程多样性和语音识别宽松度的结构化批评 | 反向 | 评测样本量小,带有编辑立场 |
| JustUseApp reviews | 独立投诉汇总 | 退款、语言覆盖和内容缺失投诉公开可见 | 反向 | 汇总质量弱于第一方应用商店数据 |
本表单独列出评测来源质量,以便留存表专注于耐久性指标。
[CU008, CU036, CU037, CU039, CU040, CU041]6.4 展示图表
07风险
7.1 法律、监管与消费者保护风险
Speak 当前并无明显的公开执法风险,但其法律面比普通语言应用更为复杂——因为产品是语音优先、AI 中介、 全球覆盖且以订阅为驱动的。Google Play 显示该应用可能与第三方共享应用活动和设备标识,也可能收集 个人及财务信息,同时提供传输加密和删除请求。Apple 给出 13+ 年龄评级,而 Google 则评为全年龄段, 使这款邀请用户对着 AI 家教自然说话的服务,对未成年用户呈现出模糊的面向姿态。这一点非常重要: 美国联邦贸易委员会(FTC)的指引指出,若服务面向 13 岁以下儿童或有实际知情的情况,COPPA 义务即行 适用;AI 法案和 GDPR 义务也可能随着 AI 系统更具实质影响或透明度降低而加重。隐私之外,账单权利 也处于碎片化状态:Apple 控制 App Store 退款,Google 施加严格的透明度和取消规则,Speak 自己的 帮助中心则对取消订阅和退款作出细致区分。因此,剩余法律风险虽可控但真实存在:一半是监管合规问题, 一半是消费者信任的配套管道问题。[CR003, CR004, CR005, CR008, CR010, CR012]
| 规则 / 案例 | 司法管辖区 | 状态 | 可能性 | 严重程度 | 缓解措施 | 残余风险 | 尽调路径 |
|---|---|---|---|---|---|---|---|
| 语音数据、面向未成年人及家长同意敞口 | 美国 / 全球 | 若 Speak 被 13 岁以下用户使用或明知收集其个人数据则为活跃风险 | 中 | 高 | Apple 和 Google 已提供年龄 / 隐私披露;Google 表示数据可删除且传输时加密 | 中高 | 请求分市场的年龄筛查逻辑、儿童数据管控、语音数据留存计划和家长同意工作流。 |
| AI 法案 / GDPR 透明度与评估义务 | 欧盟 | 自 2024 年 8 月起分阶段实施 | 中 | 高 | Speak 明确将产品定位为 AI 驱动,当前用途主要集中于辅导而非受监管的决策 | 中 | 在扩大评估或丰富语音分析之前,需将每项面向欧盟的 AI 功能映射至透明度、日志记录、DPIA/FRIA 及特殊类别数据假设。 |
| 自动续订、退款及循环计费合规 | 美国 / 欧盟 / 应用商店 | 持续进行 | 中高 | 中高 | Help-center 流程已存在,Apple / Google 各自提供取消和退款管控 | 中 | 测试实际披露页面、试用转付费通知、分应用商店的拒付率,以及各客户端中取消操作是否足够显眼。 |
| 跨境传输与第三方披露风险 | 欧盟 / 全球 | 持续进行 | 中 | 高 | 应用商店披露已涵盖部分共享、收集、加密和删除管控 | 中高 | 获取完整隐私政策、服务条款、数据处理协议、子处理商名单、标准合同条款或传输框架,以及语音转录的模型训练排除条款。 |
各行按残余严重程度排序,聚焦于对全球语音 AI 订阅应用投资者最可能影响决策的公开法律敞口。
[CR003, CR004, CR005, CR008, CR010, CR012]7.2 运营质量与平台依赖风险
在用户每次出错都立刻察觉的产品里,语音识别质量和对话延迟决定了 Speak 的运营下限。Speak 之所以自研 ASR,是因为第三方语音识别在处理口音学习者时 问题频出;这套系统现已跑在 Google Cloud Kubernetes 上的 Nvidia GPU 推理集群中。自研栈产品力更强,但运营底盘更窄——延迟、吞吐或云厂商故障会 直接拉低辅导质量。对 OpenAI 的依赖也如出一辙:Speak 公开使用 GPT-4、GPT-4o 和 Realtime API,同时承认语音转语音模型在指令跟随和精细反馈 上仍落后于文本模型。独立评测来源印证这不只是理论风险——用户投诉涉及识别延迟、漏词、噪声环境失效、耗电过快、设备发热,以及偶发的录音或进度保存 缺陷。产品口碑扎实,但运营要求极高:用户付费买的就是即时、准确、公平的体验。[CR022, CR023, CR024, CR025, CR026, CR027]
| 故障模式 | 可能性 | 严重程度 | 缓解成熟度 | 残余风险 | 未解决缺口 |
|---|---|---|---|---|---|
| 语音识别对带口音或连续语音存在漏识,影响反馈质量 | 中高 | 高 | 中——Speak 在第三方系统表现不佳后自研了 ASR | 高 | 无分语言或环境的公开准确率 / 误差预算披露。 |
| 实时角色扮演依赖语音到语音模型,Speak 坦承此类模型在细致辅导上仍弱于文本模型 | 中 | 高 | 低中——产品投入扎实,但坦承的模型局限仍存在 | 中高 | 无分功能的公开回退路由或质量阈值逻辑。 |
| 长时间会话可能导致设备过热或在嘈杂环境中质量下降 | 中 | 中 | 低——评测汇总显示该问题,但无公开遥测数据 | 中 | 无公开的硬件会话或电池性能数据。 |
| 计费和客服纠纷可能将用户好感转化为拒付和信任损失 | 中高 | 中高 | 中——已有完善的帮助中心内容 | 中高 | 无公开的退款率、拒付率或首次响应时间指标。 |
| 本地化和入门引导问题可能削弱向新语言和新市场的扩张势头 | 中 | 中 | 中等——需求旺盛,推进仍在继续 | 中等 | 无公开的上线就绪评估表或各国留存数据细分。 |
本清单聚焦于即便底层产品概念过硬、也可能损害实际学习体验的问题。
[CR023, CR024, CR025, CR026, CR027, CR036]| 依赖项 | 对手方 | 角色 | 集中度 | 失效场景 | 严重程度 | 缓释措施 | 剩余敞口 |
|---|---|---|---|---|---|---|---|
| 对话模型技术栈 | OpenAI | GPT-4、GPT-4o、Realtime API 及相关辅导功能 | 高 | API 定价、政策或可用性变化,拖累角色扮演质量、反馈效果或利润率 | 高 | Speak 自有学习引擎和部分定制 ASR,并非纯粹的封装层 | 高 |
| 推理 / ASR 基础设施 | Google Cloud + Nvidia | 定制 ASR 服务、GPU 及 Kubernetes 运行时 | 高 | 云故障、GPU 受限或推理成本飙升,都会拖慢或削弱反馈循环 | 高 | 自有模型提供一定灵活性 | 高 |
| iOS 分发与计费 | Apple App Store | 用户获取、排名及应用内订阅通道 | 中高 | 政策收紧或计费摩擦压低关键平台上的转化率、增加退款或降低曝光 | 中高 | Apple 通过标准流程处理退款和取消订阅 | 中高 |
| Android 分发与计费 | Google Play | Android 用户获取、排名、数据安全披露及订阅管理 | 中高 | 政策变化或披露问题拖慢 Android 增长或加剧信任摩擦 | 中高 | Google 提供明确的取消预期和已验证的评价机制 | 中高 |
| 私募资本支持 | 风险投资方 | 为规模扩张和模型投资提供融资与估值支撑 | 中等 | 若增长或经济指标走软,下一轮融资可能难以维持高溢价 | 高 | Speak 仍有活跃支持方和近期融资 | 中高 |
即便产品实力出众,若计费、AI 质量、基础设施和融资都集中于少数对手方,公司也会变得脆弱。
[CR002, CR022, CR023, CR026, CR029, CR033]7.3 财务模型与执行风险
Speak 的财务风险不在于显性困境,而在于一款高运营成本产品背后不透明的经济模型。公开数据呈现强劲势头:2024 年中用户超 1000 万、覆盖 40 余个 国家、2024 年底 Series C 估值达 10 亿美元,OpenAI、Khosla、Accel、Y Combinator 持续跟投。但同样是公开信息——Speak 承认部分功能在当前 成本下得等模型价格下降才划算,独立报道也指出自研大模型野心烧钱。这意味着毛利率、每活跃学习者推理成本、退款漏损和支持成本至关重要,而这些数据均 未公开。NicheMetric 提供了应用营收和下载量的方向性估算,但精度不足以回答核保问题。执行风险进一步加剧不确定性:公司在扩语种、招人、拓产品范围 的同时,完成了从韩国到 20 余个、再到 40 余个国家的跃升。正向评价说明这步棋走对了,但对更多语种和更好支持的持续呼声,暗示产品背后的运营体系 或许还在追赶需求。[CR002, CR029, CR030, CR031, CR032, CR033]
| 角色 / 职能 | 依赖或缺口 | 可能性 | 严重程度 | 缓释措施 | 尽调路径 |
|---|---|---|---|---|---|
| ML / ASR / LLM 工程 | 定制 ASR 加上 OpenAI 驱动的辅导功能,需要稀缺的系统和模型人才 | 中等 | 高 | 近期融资和持续招聘表明公司仍在加码技术栈投入 | 索取组织架构图、人员流失率、值班设计,以及 ASR 与角色扮演功能的供应商归属图。 |
| 客服与计费运营 | Apple、Google 及直接网页支付流程带来多语言客服复杂性 | 中高 | 高 | 帮助中心覆盖面广且内容及时 | 审查首响时间、退单处理、升级处理手册及本地化覆盖情况。 |
| 内容与本地化运营 | 语言和国家数量增多,加大了课程质量保障、合规和入门流程的复杂度 | 中等 | 高 | 用户好感度高,多国扩张已验证可行 | 按语言、地区和应用商店渠道审查留存和退款队列数据。 |
| 企业 / 商业产品线 | Speak for Business 将支持、合同和可靠性要求推至消费者应用之外 | 中等 | 中高 | 公开证据显示商业版本已上线 | 索取企业销售管线、SLA 承诺、安全问卷回复及客服人员配置。 |
本表按规模扩张复杂度超越现有公开运营体系的程度排列执行风险。
[CR033, CR048, CR049, CR052, CR053]7.4 缓解措施、监控指标与逻辑打碎触发因素
对 Speak 来说,利好在于产品已有真实用户热爱、大规模采用和部分客户支持与技术透明度的证明——这降低了主要风险是纯隐患的概率。更现实的核保问题 是:这些缓解措施能否与公司野心同步成熟?高评分和积极的口语反馈有帮助,加密与删除控制有帮助,有据可查的取消与退款流程有帮助。但这些都无法替代 完整的隐私政策、企业级可靠性承诺或硬性单元经济证据。最重要的监控指标很实际:应用商店评论中识别或计费问题的聚集、退单与退款率、语音数据留存和 使用控制不清晰,以及任何迫使产品降级或利润重置的 OpenAI 或云成本变化。对投资人而言,单次差评周不会打碎投资逻辑;打碎投资逻辑的,是反复出现的 运营投诉、合规漏洞或缺失的经济数据,开始压过一款目前凭借体验胜出的产品。[CR005, CR008, CR010, CR013, CR022, CR033]
| 风险 | 可监控触发器 | 阈值 / 事件 | 行动含义 |
|---|---|---|---|
| 订阅信任 / 计费摩擦 | 退款申请、退单及与计费相关的评价激增 | 定价或试用流程调整后数周内出现显著跳升 | 暂停激进获客,在加大投放前重新梳理应用商店及入门引导的披露内容。 |
| 识别质量 | 大量评价提及漏词、延迟或发音窗口过短 | 问题持续出现于多个主要语言或平台版本 | 在准确率和节奏恢复前暂停语言扩张。 |
| 语音数据 / 儿童数据合规 | 无法向尽调方或监管机构提供数据留存、删除或年龄筛查管控证明 | 缺少 DPIA / 留存方案,或收到监管 / 平台通知 | 在合规文件齐备前下调承保评级。 |
| OpenAI / 模型依赖 | API 价格实质性上涨、延迟恶化或政策约束 | 利润率持续受损或实时角色扮演体验下降 | 重置估值假设,要求提供备用方案 / 路由切换证明。 |
| 云 / 推理集中 | 反复出现 Sev-1 级推理或语音服务中断 | 发生超过一次重大故障且无可信的容灾切换证明 | 在承保溢价增长前,要求提交多地区或多供应商韧性方案。 |
| 融资 / 估值支撑 | 新一轮融资或二级市场流程暗示平价 / 下调,而缺乏指标支撑 | 无法证明留存率、利润率和退款纪律具有持续性 | 将当前估值视为脆弱,避免在经济指标核实前提前押注。 |
这些触发器将宏观尽调叙述转化为投资人可落地的监控框架。
[CR005, CR010, CR014, CR023, CR026, CR033]Speak 最高的残余风险集中在 AI/语音监管、应用商店计费、模型/云依赖与高溢价消费体验相互交叉的地带。
[CR003, CR005, CR008, CR010, CR016, CR019]Speak 的主要风险通过几条主线传导:合规、计费信任、模型质量、利润率和估值支撑。
[CR003, CR008, CR010, CR022, CR023, CR024]Speak 的消费级 AI 产品依赖一组紧凑的外部基础设施:模型供应、云推理、分发渠道、计费和支持处理。
[CR001, CR002, CR009, CR022, CR026, CR033]7.5 附件
08估值
8.1 投资建议与定价纪律
Speak 的公开运营证据足以支撑认真尽调,但不足以支撑以最近公开的 10 亿美元估值干净买入。官方披露显示估值从 2024 年 6 月的 5 亿美元跳至 2024 年 12 月的 10 亿美元,最新报道指向超 1 亿美元年化收入、1500 万次下载、1000 万以上 Google Play 安装量以及真实的企业业务。与大多数 消费教育科技初创公司相比,这是明显更强的证据。问题在于可参考的价格锚点依然隐含高个位数到约 10 倍的收入乘数,远高于当前公开教育和订阅学习 可比公司。Duolingo 是唯一兼具类似增速和产品质量的上市标的,但按当前市场数据其 EV/收入也只在 4 倍左右。公开估值早于最新收入里程碑,股权 结构表、优先股堆叠和留存结构亦未披露,审慎结论是:继续研究,置信度中,风险高,估值偏高。只有当深入尽调揭示比现有公开证据更强的经常性经济数据, 或入场价格大幅低于最近公开标记,进入才具吸引力。[CV001, CV006, CV012, CV015, CV017, CV021]
| 维度 | 评估 | 证据支持的理由 | 决策含义 |
|---|---|---|---|
| 建议 | research-more | 产品吸引力真实存在,但公开证据尚不足以支撑在上轮 10 亿美元估值下进行价格敏感型承保。 | 仅在获取完整私募尽调材料后方可推进。 |
| 可信度 | 中等 | 融资、营收、定价、应用吸引力和可比公司数据均可见,但留存率和股权结构经济性尚不透明。 | 将公开信息视为筛选工具,而非投委会可直接使用的结论。 |
| 风险评级 | 高 | 结果取决于持续的溢价增长,以及尚未披露的稀释比例和优先清算条款。 | 在做出任何意向书决策前,明确为下行情景建模。 |
| 估值立场 | stretched | 隐含私募倍数高于公开教育科技可比公司区间,需要更优的增长质量来支撑溢价。 | 寻求价格让步或异常清晰的经济条款。 |
| 目标承保 | ≥3x 总 MOIC(4-5 年) | 按上轮公开估值,该目标很可能需要比现有公开证据所证明的更快的规模扩张。 | 优先在低于上一轮的价格入场,或要求强下行保护条款。 |
目标回报 / 持有期框架是承保纪律,而非来源于市场报价;之所以纳入,是因为第 08 章需要决策含义。
[CV006, CV012, CV050, CV051, CV053, CV057]从已观察到的增长势头和市场顺风出发,到估值谨慎和"继续研究"建议的推导路径。
[CV012, CV017, CV021, CV041, CV042, CV051]投委会风格的 0-10 评分卡,将各章节证据转化为决策因素。
评分为分析师序数判断(0-10),综合自引用论断。非公司来源 KPI,仅用于汇总各章节维度的投资质量。
[CV012, CV017, CV021, CV041, CV042, CV046]8.2 情景区间与可比公司框架
公开市场锚点表明 Speak 值一个高于困境教育科技标的的溢价,但溢价必须靠异常持久的增长和变现来挣到。Duolingo 增速健康、利润为正、市值数十亿, EV/收入约 4 倍;Coursera、Udemy 和 Chegg 远低于 1 倍 EV/收入。这个落差很重要——Speak 仍是私有公司,投资人进入时的披露程度不如上述任何 一家。上行情景仍有意义:AI 辅导和在线语言学习都是体量大、仍在增长的市场,Speak 的应用排名、定价和企业牵引力表明它在打造一个比普通词汇游戏 更严肃的学习产品。不过,MMR 的市场研究标记了免费增值竞争和获客成本(CAC)压力,Oliver Wyman 也展示了当市场质疑座位增长、护城河持久性或 收入质量时,AI 软件倍数能多快重新定价。这就是情景框架采用保守公开可比区间而非外推开放式 AI 溢价的原因。悲观情景压缩至 4 亿到 6.5 亿美元, 基准情景维持在 8 亿到 10 亿美元,乐观情景要求收入远超 1.5 亿美元、企业转化更广且应用商店势头延续。[CV025, CV029, CV033, CV037, CV041, CV042]
| 视角 | 论点 | 反论点 | 改变判断的条件 |
|---|---|---|---|
| 需求 | 庞大的 AI 辅导和语言学习市场,为类目持续增长提供支撑。 | 大市场并不能阻止获客成本膨胀,也无法抵御免费 / 免费增值工具的竞争。 | 按地区呈现队列级别的留存率和高效获客数据。 |
| 商业化 | 应用商店定价及超 1 亿美元年化营收表明消费者愿意为优质口语产品付费。 | 公开证据不能说明转化可持续性,也无法证明营收是否过度依赖消费者端。 | 提供转化漏斗、回收周期和续订指标。 |
| 企业业务 | 2024 年 12 月 200+ 客户,2025 年末约 500 家雇主,暗示 B2B 扩张期权价值。 | 客户数量不等于持久 ARR、扩张能力或毛利润证明。 | 披露企业 ARR、续约率和头部客户集中度。 |
| 相对估值 | 若 Speak 更接近类目领导者而非普通应用,对弱势公开教育科技同行的溢价就有正当性。 | 当前可比公司区间仍暗示 10 亿美元估值已消化大量未来执行兑现。 | 若以相近价格完成新一轮融资并改善信息披露,将提升判断的可信度。 |
| 情景 | 假设条件 | 估值 / 回报逻辑 | 关键风险 | 概率信号 |
|---|---|---|---|---|
| 悲观 | 增长放缓,应用势头减弱,投资人按公开教育科技区间为 Speak 定价。 | 估值 4 亿—6.5 亿美元;若以 10 亿美元入场,在稀释和优先权生效前已可能亏损。 | 倍数压缩、付费墙摩擦、企业转化偏弱。 | 若应用商店势头消退或板块情绪再次走弱,则更可能发生。 |
| 基准 | 营收规模维持在当前运营轨道附近,企业业务扩张但仍未经证实,对公开同行的溢价收窄但不消失。 | 估值 8 亿—10 亿美元;按上一轮入场,若条款不够优越,上行空间有限。 | 业务结构 / 留存不透明、融资需求、难以守住溢价。 | 与现有公开证据最为吻合。 |
| 乐观 | 营收明显超过 1.5 亿美元,企业渗透拓宽,Speak 保持高溢价应用势头并披露更强劲的经济指标。 | 估值 11 亿—14 亿美元;上行空间存在,但仍依赖私有数据确认商业化的可持续性。 | 执行失误、竞品模仿、续约质量不及预期。 | 需要公开信息来源目前无法提供的证据。 |
情景区间是基于已公开营收里程碑、当前公开可比公司及 2026 年软件估值倍数风险评论的分析估算;不包含未披露优先权的影响。
[CV012, CV021, CV025, CV029, CV033, CV037]| 可比公司 | 指标 | 倍数 / 估值 / 状态 | 相关性 | 局限性 |
|---|---|---|---|---|
| Duolingo | TTM 营收 / EV/营收 / 市值 | 营收 10.4 亿美元;EV/营收 4.01 倍;市值 51.5 亿美元 | 优质语言学习软件的最佳公开基准。 | 成熟度、盈利能力和信息披露均远超 Speak。 |
| Coursera | TTM 营收 / EV/营收 / 市值 | 营收 7.898 亿美元;EV/营收 0.42 倍;市值 9.8 亿美元 | 展示增长和利润率偏弱时,综合在线学习平台的定价水平。 | 业务结构比口语主导的消费者语言学习更为宽泛。 |
| Udemy | TTM 营收 / EV/营收 / 市值 | 营收 7.739 亿美元;EV/营收 0.38 倍;市值 6.8 亿美元 | 适用于增长较慢订阅学习的消费者 / 创作者教育基准。 | 市场经济模式与 Speak 自有课程模式不同。 |
| Chegg | TTM 营收 / EV/营收 / 市值 | 营收 3.769 亿美元;EV/营收 0.32 倍;市值 1.2 亿美元 | 教育资产在产品 / 护城河受损后可能跌至的下行锚点。 | Chegg 是反面案例,非目标同行。 |
当前用于价格纪律的公开锚点样本;私募 AI 辅导可比公司仍过于不透明,无法从公开来源详尽列举。
[CV025, CV027, CV028, CV029, CV031, CV032]基于公开证据,展示不同收入 / 倍数组合下的企业价值示意结果。
数值为美元百万级示意数字,源自公开报道的年化收入超 $100M 里程碑加上可观察的公开可比倍数,并四舍五入为决策参考锚点。这些数字不代表公司指引或交易报价。
[CV012, CV025, CV029, CV033, CV037, CV050]仅基于公开证据的 Speak 悲观 / 基准 / 乐观估值区间。
数值为美元百万级示意数字,基于公开收入里程碑、当前上市同类公司倍数及 2026 年软件倍数风险评论。不含未披露的稀释、债务或清算优先权影响,投资者回报区间因此仍不完整。
[CV050, CV051, CV054, CV055, CV056]8.3 升级触发因素、逻辑打碎条件与最终问题清单
从继续研究到买入的路径清晰,但要求证据充分。Speak 需要证明企业牵引力不只是 logo 数量,而是有扎实续约质量的经常性扩展收入;还需要证明消费者 规模正转化为留存和队列经济,足以为高于公开教育科技的溢价倍数提供支撑。主要的逻辑打碎条件清晰可观测:应用商店势头逆转、消费者付费墙摩擦加剧、 企业采用无法转化为披露的 ARR 支撑,或更广泛的 AI 软件倍数再次下移。监管与市场结构风险同样重要——生成式 AI 平台可能向少数关键模型、数据或 算力持有者集中,压缩应用层公司的护城河。市场在 Speak 完成足够经济披露之前停止奖励 AI 叙事,融资窗口也可能迅速关闭。换句话说,Speak 看起来 是一门可信的生意,但核保缺口在于回报架构而非产品存在本身。没有股权结构表、单元经济和分部构成,投资人仍在为叙事溢价定价;有了这些,公司可能值 得观察乃至更高评级;没有,正确姿态是纪律性尽调而非确信定价。升级路径因此在今天就是具体的:证明分部级留存,展示企业续约质量,并证明溢价定价在 竞争压力下守得住。[CV009, CV013, CV022, CV046, CV047, CV053]
| 触发器 | 阈值 | 对论点的传导 | 行动含义 |
|---|---|---|---|
| 应用势头消退 | 核心市场下载量 / 排名数据连续数月失去高溢价定位 | 削弱 Speak 仍在以高溢价赢得消费者注意力的证据。 | 将基准情景下调至公开同行区间。 |
| 企业证明停滞 | 尽管雇主覆盖持续增长,却无披露的 ARR、续约或扩张数据 | 将企业故事降格为叙事附加项,而非支撑估值的第二增长引擎。 | 不为 B2B 期权价值支付溢价。 |
| 板块倍数压缩加剧 | AI / 软件风险再次重新定价,高溢价订阅标的进一步去评级 | 即便运营保持稳健,估值上限也会收窄。 | 要求进一步折价于上一轮,或暂时观望。 |
| 条款对投资人不友好 | 2024-2026 融资堆栈中出现重度优先权、棘轮条款或隐性稀释 | 即便企业价值维持在基准情景附近,也可能抹去回报。 | 放弃或坚持保护性入场条款。 |
| 主题 | 缺失证据 | 重要性 | 负责方或尽调路径 |
|---|---|---|---|
| 股权结构 / 优先权 | 无公开的清算优先权瀑布、期权池或优先清算条款细节 | 回报测算可能与企业价值增长产生重大偏差。 | 公司法律顾问 + 领投方数据室。 |
| 业务分部结构 | 无公开的消费者与企业营收拆分或地理分布数据 | 高估值依赖营收质量与多元化。 | 财务团队营收拆分表及董事会材料。 |
| 队列经济学 | 无公开的转化率、留存率或 CAC 回收期披露 | 判断营收能否高效复利增长的必要依据。 | 按队列导出增长 / 财务分析数据。 |
| 企业质量 | 200-500 家雇主覆盖中,无公开的合同期限、续约率或扩张数据 | 验证乐观情景持续性和下行保护的必要依据。 | 销售运营管线审查加头部客户尽调访谈。 |
8.4 附件
免责声明
本报告为基于公开证据的尽调快照,不构成投资建议。重要的财务、法律、技术和合同事实尚未公开,在作出任何投资决策前,应直接与管理层及原始文件核实。
证据索引
| 编号 | 陈述 | 可信度 | 来源 |
|---|---|---|---|
| CO001 | Official and third-party company profiles place Speak (Speakeasy Labs) founding in 2016. | 中 | SO007, SO012, SO016 |
| CO002 | Speak is headquartered in San Francisco, California. | 中 | SO009, SO011, SO016, SO013 |
| CO003 | Speak markets itself as an AI language tutor centered on getting learners speaking out loud with instant feedback. | 中 | SO001, SO004 |
| CO004 | By May 2026, Speak publicly promoted courses for French, Spanish, English, Korean, Italian, and Japanese. | 中 | SO001, SO004, SO008 |
| CO005 | Speak homepage marketing in May 2026 claimed 15M+ downloads and a 4.8 rating. | 中 | SO001 |
| CO006 | The iOS App Store listing showed 44K ratings and a 4.8 score on 2026-05-05. | 中 | SO004 |
| CO007 | The Google Play listing showed 112K reviews and a 4.7 score on 2026-05-05. | 中 | SO005 |
| CO008 | Speak announced a $20M Series B-3 financing at a $500M valuation on 2024-06-18. | 高 | SO006, SO009 |
| CO009 | Speak announced a $78M Series C financing at a $1B valuation on 2024-12-10. | 高 | SO007, SO010, SO011 |
| CO010 | Speak said total funding reached $162M after the Series C round. | 中 | SO007, SO010 |
| CO011 | Accel led Speak’s Series C, with OpenAI Startup Fund, Khosla Ventures, and Y Combinator among participating investors. | 高 | SO007, SO010, SO011 |
| CO012 | HolonIQ listed Speak as joining the global EdTech unicorn list in December 2024 at a $1B valuation. | 高 | SO015, SO007 |
| CO013 | By June 2024, Speak said it had more than 10 million learners in over 40 countries. | 中 | SO006, SO009 |
| CO014 | TechCrunch reported Speak had a 75-person workforce across San Francisco, Seoul, Tokyo, and Ljubljana in June 2024. | 中 | SO009 |
| CO015 | Speak’s careers page names Seoul, Ljubljana, and San Francisco offices and presents the company as globally distributed. | 中 | SO003 |
| CO016 | The careers page also contains a historical snapshot claiming a 60-person team and more than $60M raised. | 低 | SO003 |
| CO017 | Speak said it launched Speak for Business between the June and December 2024 funding rounds. | 中 | SO007, SO002 |
| CO018 | The current B2B page says 200+ brands rely on Speak for Business. | 中 | SO002 |
| CO019 | Speak’s Series C post said its enterprise offering had more than 200 customers and an 85% employee adoption rate. | 中 | SO007 |
| CO020 | Speak reported that users had already spoken more than one billion sentences in 2024. | 中 | SO007 |
| CO021 | Speak’s June 2025 product update announced four new languages for English speakers—French, Japanese, Korean, and Italian—after the earlier Spanish release. | 中 | SO008 |
| CO022 | The June 2025 post says Speak started by teaching English in Korea, Japan, and Taiwan and had more than 15 million learners globally. | 中 | SO008 |
| CO023 | Speakeasy Labs filed a Form D on 2024-12-11 for a $77,699,277 equity offering under Rule 506(b), with first sale on 2024-11-13. | 高 | SO013, SO014 |
| CO024 | The filing page lists Connor Zwick, Andrew Hsu, Colton Gyulay, Alex Berkenkamp, and Ben Quazzo among related persons and shows a 100 Pine Street San Francisco address. | 中 | SO013 |
| CO025 | The public filing list shows additional Speakeasy Labs Form D filings dated 2024-08-12 and 2023-10-17. | 中 | SO014 |
| CO026 | Speak’s June 2024 announcement said learners speak about 1,000 times on average in their first week. | 中 | SO006 |
| CO027 | Speak’s June 2024 announcement said an updated speech recognition model reduced word error rate by over 60% and improved speed by 20% versus existing commercial systems. | 中 | SO006 |
| CO028 | Speak’s December 2024 announcement said it created more than 25 million personalized lessons during 2024. | 中 | SO007 |
| CO029 | The founders publicly identified in late-2024 coverage are Connor Zwick (CEO/co-founder) and Andrew Hsu (CTO/co-founder). | 高 | SO010, SO012 |
| CO030 | Accel partner Ben Quazzo joined Speak’s board as part of the Series C round. | 高 | SO007, SO010 |
| CO031 | TechCrunch’s June 2024 article described Speak as launched in 2014, conflicting with the company’s 2016 founding references elsewhere. | 低 | SO009 |
| CO032 | Android Police concluded in February 2026 that Speak’s voice-driven learning can give users a false sense of mastery because it misses basic pronunciation errors. | 中 | SO017, SO005 |
| CO033 | JustUseApp’s review aggregation labeled overall customer experience 67.1% negative and highlighted recurring complaints about billing, lag, and recognition. | 低 | SO018 |
| CO034 | Languatalk’s 2026 review said Speak is polished for beginners but feedback depth, lesson variety, and premium tier clarity remain weak points. | 中 | SO019 |
| CO035 | GetLatka reports Speak reached $100M revenue and 253 employees in 2025, but those figures are not corroborated by official Speak disclosures. | 低 | SO016 |
| CO036 | The current official homepage and app listings show Speak has expanded from an English-only product into a broader multi-language learning app. | 高 | SO001, SO004, SO008 |
| CO037 | Speak positions English learning as the initial wedge but now frames the product as a broader language tutoring platform for consumers and employers. | 中 | SO002, SO007, SO008 |
| CM001 | Speak’s June 2024 financing announcement described the addressable market as a $100B+ online and in-person language learning market. | 中 | SM002 |
| CM002 | TechCrunch reported Speak initially focused on English because it is the world’s most popular language for learning. | 中 | SM013 |
| CM003 | TechCrunch quoted Speak’s CEO saying roughly 1.5 billion people are trying to learn English. | 中 | SM013 |
| CM004 | Preply’s 2026 report says English is the most learned language because of its global role in business and education. | 中 | SM011 |
| CM005 | Preply states English has about 1.5 billion total speakers worldwide. | 中 | SM011 |
| CM006 | Technavio says the digital English language learning market will grow by USD 39.46B from 2024 to 2029 at a 24.5% CAGR. | 中 | SM005 |
| CM007 | Technavio identifies APAC as the largest regional market and says it will contribute 39% of forecast growth. | 中 | SM005 |
| CM008 | Technavio says increased flexibility from digital language courses is a primary growth driver for the market. | 中 | SM005 |
| CM009 | Technavio says corporate non-academic learners are a major digital English customer segment, with roughly 30% opting for digital courses. | 中 | SM005 |
| CM010 | MarketsandMarkets projects the broader AI-in-education market to grow from USD 2.21B in 2024 to USD 5.82B by 2030 at a 17.5% CAGR. | 中 | SM006 |
| CM011 | MarketsandMarkets says North America held a 43% share of the AI-in-education market in 2024. | 中 | SM006 |
| CM012 | MarketsandMarkets says personalized learning and content management accounted for 34.5% of the AI-in-education market in 2024. | 中 | SM006 |
| CM013 | The same report names Duolingo and ELSA Speak among notable AI-in-education players. | 中 | SM006 |
| CM014 | Stanford HAI reports four out of five U.S. high school and college students now use AI for schoolwork. | 中 | SM007 |
| CM015 | WEF says AI can automate or augment up to 20% of educator clerical tasks. | 中 | SM008 |
| CM016 | WEF also says equitable access, data privacy, bias, and teacher displacement are major constraints on AI-in-education adoption. | 中 | SM008 |
| CM017 | World Bank says low- and middle-income countries face steep challenges adapting or deploying AI at scale. | 中 | SM009 |
| CM018 | World Bank frames the foundations of scalable AI adoption as connectivity, compute, context, and competency. | 中 | SM009 |
| CM019 | The 2021 systematic review says most AI language tools used machine learning and natural language processing for error identification, feedback, and assessment. | 中 | SM010 |
| CM020 | The same review concludes AI language tools improved learner abilities but raised privacy and teacher-preparation concerns. | 中 | SM010 |
| CM021 | Preply estimates the global online language learning market reaches about $115B by the end of 2025. | 低 | SM011 |
| CM022 | Preply estimates the English learning segment is worth about $43.51B in 2025 and growing around 22% annually. | 低 | SM011 |
| CM023 | Speak’s current product spans both consumer self-serve subscriptions and employer-sponsored learning through Speak for Business. | 中 | SM001, SM003 |
| CM024 | Speak’s Series C post says English learning is industry agnostic and its business product already had 200+ customers. | 中 | SM003 |
| CM025 | TechCrunch reported Speak’s consumer list price as $20/month or $99/year in 2024. | 高 | SM014, SM013 |
| CM026 | Duolingo’s 2024 annual report says it serves more than 100M monthly active users across 40+ languages and only about 9% of MAUs are paid subscribers. | 中 | SM017 |
| CM027 | Duolingo’s 2024 annual report says the app is the top-grossing education app globally on both Apple and Google app stores. | 中 | SM017 |
| CM028 | ELSA positions itself as a specialized AI English speaking coach with 18M+ downloads and 460K+ ratings on its website. | 中 | SM018 |
| CM029 | ELSA’s App Store listing shows 109K ratings and monthly and yearly memberships, reinforcing a premium subscription model for speaking practice. | 中 | SM019 |
| CM030 | Cambly competes with Speak from a different labor model: real conversations with native speakers available 24/7 rather than AI-only tutoring. | 中 | SM020 |
| CM031 | Busuu emphasizes community feedback from native speakers and says it has 120M+ registered users. | 中 | SM022 |
| CM032 | Busuu’s App Store listing positions the product as community-driven learning rather than AI-first conversation tutoring. | 中 | SM021 |
| CM033 | Babbel’s app-store surfaces emphasize expert-built structured lessons, with 25M subscriptions sold and 50M+ Google Play downloads. | 中 | SM023, SM024 |
| CM034 | Praktika markets an AI-tutor model with 20M+ learners and a much lower-cost alternative to private human tutors. | 中 | SM025 |
| CM035 | Android Police argued Speak can give learners a false sense of mastery because its pronunciation scoring is overly lenient. | 中 | SM015 |
| CM036 | Languatalk found Speak strongest for early learners but weaker for serious learners who need deeper feedback and broader lesson variety. | 中 | SM016 |
| CM037 | Speak’s June 2025 post says summer travel increases language-learning interest and ties learning motivation to travel identity. | 中 | SM004 |
| CM038 | Preply says progress still depends on access, indicating affordability and digital infrastructure remain barriers even as demand rises. | 中 | SM011 |
| CP001 | Speak’s core competitive claim is speaking-first AI tutoring rather than broad textbook-style language learning. | 中 | SP001, SP002 |
| CP002 | By 2026 Speak was no longer English-only: official surfaces advertise six live learning tracks and a broader multi-language ambition. | 中 | SP001, SP033 |
| CP003 | Speak’s current consumer price anchor remains about $20/month or $99/year in public reporting. | 高 | SP005, SP006 |
| CP004 | Speak for Business claims 200+ customers or brands and gives Speak a second distribution path beyond direct-to-consumer subscriptions. | 中 | SP004, SP023 |
| CP005 | Speak has materially smaller public scale than Duolingo, which reported 100M+ MAUs and 40+ languages in its 2024 annual report. | 中 | SP009, SP001 |
| CP006 | Duolingo’s freemium model relies on huge free-user scale with only about 9% of MAUs paying, making it a powerful low-cost substitute for Speak. | 中 | SP009 |
| CP007 | Duolingo is the top-grossing education app globally on Apple and Google app stores, reinforcing its distribution advantage over smaller challengers. | 中 | SP009 |
| CP008 | ELSA is a direct specialized speaking competitor focused specifically on English pronunciation and conversation rather than broad multilingual learning. | 中 | SP010, SP011 |
| CP009 | ELSA publicly claims 18M+ downloads and 460K+ ratings on its website. | 中 | SP010 |
| CP010 | ELSA’s App Store listing shows 109K ratings and monthly/yearly memberships, signaling a scaled subscription business in the same English-speaking niche. | 中 | SP011 |
| CP011 | Cambly competes using human native-speaker conversations available 24/7, making it a live-tutor substitute rather than an AI-only product. | 中 | SP012 |
| CP012 | Busuu competes with a community-feedback model and says it has more than 120M registered users. | 中 | SP013, SP014 |
| CP013 | Busuu’s Play listing shows 50M+ downloads and 1.13M reviews, reflecting broad consumer reach even without an AI-first pitch. | 中 | SP015 |
| CP014 | Babbel competes from the structured-course end of the market and says it has sold 25M subscriptions. | 中 | SP016 |
| CP015 | Babbel’s Play listing shows 50M+ downloads and 1.12M reviews, giving it much larger installed-base distribution than Speak. | 中 | SP017 |
| CP016 | Praktika is an AI-tutor competitor claiming 20M+ learners and a roughly $8/month price point positioned against human tutors. | 中 | SP018 |
| CP017 | Loora competes as an always-available AI English tutor with business-English and real-time feedback positioning. | 中 | SP019 |
| CP018 | MarketsandMarkets names both Duolingo and ELSA as AI-in-education players, suggesting the competitive set spans broad platforms and specialized speaking tools. | 中 | SP020 |
| CP019 | HolonIQ’s January 2026 addition of Preply to the EdTech unicorn list signals that language learning remains a venture-funded, still-fragmenting category. | 中 | SP021 |
| CP020 | Speak’s strongest relative differentiation is still its speaking-first UX and AI tutoring focus, not raw scale. | 中 | SP001, SP005, SP007 |
| CP021 | Speak’s multi-language expansion narrows one historical weakness against incumbents, but its public breadth still trails broader platforms like Duolingo, Busuu, and Babbel. | 中 | SP001, SP009, SP013, SP016, SP033 |
| CP022 | Cambly’s human-tutor model means Speak is not only fighting apps; it is also fighting live conversation as the trusted premium substitute. | 中 | SP012, SP005 |
| CP023 | Busuu and Babbel compete on structure, community, and breadth rather than pure AI conversation, giving budget-conscious users alternatives to Speak. | 中 | SP013, SP016, SP017 |
| CP024 | Praktika and Loora show that AI-speaking competition is no longer niche; multiple challengers now market conversation practice as a human-tutor replacement. | 中 | SP018, SP019 |
| CP025 | Android Police described Speak as heavily inspired by Duolingo, reducing the novelty moat around its course structure and gamification. | 中 | SP007 |
| CP026 | Android Police also found Speak’s pronunciation scoring too forgiving, which weakens one of the product claims that should be most defensible versus broad language apps. | 中 | SP007 |
| CP027 | Languatalk judged Speak polished for beginners but weaker for serious learners who want deeper feedback, better review loops, and richer advanced practice. | 中 | SP008 |
| CP028 | JustUseApp’s complaint aggregation indicates billing, refund, and support issues can damage trust even when the top-line app rating remains strong. | 中 | SP024, SP002, SP003 |
| CP029 | ELSA’s and Speak’s app-store footprints are much closer to each other than either is to Duolingo’s scale, making ELSA a more relevant specialized benchmark than Duolingo alone. | 中 | SP002, SP003, SP010, SP011, SP009 |
| CP030 | Speak’s B2B motion is a real differentiator versus most consumer-only rivals, but the public record still lacks seat counts, customer names, and renewal rates. | 中 | SP004, SP023 |
| CP031 | Preply’s 2026 report reinforces that English remains the most learned language because of business and education demand, which favors all major competitors rather than Speak alone. | 中 | SP022 |
| CP032 | Babbel and Busuu both emphasize more structured pedagogy and community reinforcement than Speak’s free-conversation-centered positioning. | 中 | SP013, SP014, SP016 |
| CP033 | ELSA emphasizes pronunciation, role-plays, and bilingual support, making it especially strong where buyers want more explicit correction than Speak’s reviews suggest it delivers. | 中 | SP010, SP011, SP007 |
| CP034 | Praktika uses a lower-price, AI-tutor framing that could pressure Speak if speaking practice becomes commoditized. | 中 | SP018, SP005 |
| CP035 | Duolingo’s enormous free-user funnel and top-grossing status mean it can pressure smaller players on both acquisition cost and user expectations. | 中 | SP009 |
| CP036 | Cambly, Busuu, Babbel, Duolingo, ELSA, Praktika, and Loora together cover live tutoring, community correction, structured coursework, and AI conversation—meaning buyers have multiple non-Speak ways to solve the same job. | 中 | SP009, SP010, SP012, SP013, SP016, SP018, SP019 |
| CP037 | There is no reviewed public source giving a clean market-share ranking for Speak versus these rivals, which is itself a diligence gap. | 低 | |
| CP038 | Duolingo’s official homepage still leads with a free value proposition and very broad course catalog, reinforcing its role as the default low-cost substitute. | 中 | SP025, SP032 |
| CP039 | Cambly’s own pricing page highlights one-on-one and Pro tutoring plans with native speakers, showing that the human-tutor substitute is productized rather than bespoke. | 中 | SP027, SP028 |
| CP040 | Preply and italki extend the substitute set beyond apps into tutor marketplaces, while HelloTalk extends it into peer-to-peer exchange, increasing buyer choice around the same fluency job. | 中 | SP029, SP030, SP031 |
| CP041 | The most durable moat visible publicly is distribution into business accounts plus speaking-first UX, but that moat looks moderate rather than dominant because rivals match on either scale or tutor quality. | 中 | SP004, SP009, SP012, SP018 |
| CI001 | Speak’s public monetization architecture is hybrid: consumer subscriptions in the app stores plus employer-facing contracts through Speak for Business. | 中 | SI002, SI004 |
| CI002 | Speak’s public consumer price anchor is consistently described as about $20 per month or $99 per year. | 高 | SI007, SI008 |
| CI003 | The App Store listing corroborates that Speak sells monthly and annual auto-renewing subscriptions rather than a pure one-time purchase. | 中 | SI002 |
| CI004 | Speak for Business publicly claims 200+ brands rely on the product, indicating a real but still sparsely disclosed B2B revenue line. | 中 | SI004, SI006 |
| CI005 | Speak has strong public engagement proxies—15M+ downloads, 44K App Store ratings, and 112K Google Play reviews—but none of these disclose paying subscribers or ARR. | 中 | SI001, SI002, SI003 |
| CI006 | TechCrunch reported users spend roughly 10-20 minutes per day in Speak, which supports engagement but not monetized retention. | 中 | SI008 |
| CI007 | There is no reviewed public source disclosing Speak revenue, ARR, gross margin, net retention, or cash balance. | 低 | |
| CI008 | Recent officially announced fundraising totals roughly $98M gross across the June 2024 Series B-3 and December 2024 Series C rounds. | 高 | SI005, SI006 |
| CI009 | The Series B-3 announcement attached a $500M valuation to Speak, and the Series C announcement attached a $1B valuation six months later. | 高 | SI005, SI006 |
| CI010 | The Speakeasy Labs Form D reviewed via FilingFlow lists a total offering amount of $77.7M with first sale on November 13, 2024, broadly matching the announced Series C size. | 中 | SI009 |
| CI011 | The SEC-filing list implies recurring private financing activity, but it does not disclose cash still on hand after those raises. | 中 | SI010 |
| CI012 | Because Speak does not publish balance-sheet data, any runway assessment remains scenario-based rather than underwritten from public statements. | 中 | SI005, SI006, SI009 |
| CI013 | Speak’s revenue mix between consumer subscriptions and enterprise contracts is not publicly quantified. | 低 | |
| CI014 | Enterprise pricing, seat counts, contract length, and renewal data for Speak for Business are absent from reviewed public sources. | 低 | |
| CI015 | Revenue quality cannot be judged confidently from list price alone because public sources do not reveal free-trial conversion, refund rate, churn, or realized discounts. | 中 | SI002, SI007, SI008 |
| CI016 | Public complaint aggregation points to refund and support friction, which could weigh on net consumer revenue quality even if headline app ratings stay high. | 中 | SI024, SI002, SI003 |
| CI017 | Duolingo’s 2024 annual report is the clearest public-company proxy that language-learning at scale monetizes through a freemium funnel with only a minority of MAUs paying. | 中 | SI011 |
| CI018 | Duolingo’s official homepage still leads with a free value proposition, reinforcing how strong the low-price anchor is for the category. | 中 | SI023 |
| CI019 | Compared with Speak’s monthly subscription, Cambly monetizes live tutoring in higher-value lesson units rather than flat app access. | 中 | SI012, SI002 |
| CI020 | italki monetizes as a tutor marketplace: its teachers page exposes thousands of English tutors and visible trial prices as low as USD 5. | 中 | SI017 |
| CI021 | HelloTalk’s messaging emphasizes free language exchange at 70M+ registered users and 260+ languages, creating a zero-price substitute for casual practice. | 中 | SI018 |
| CI022 | Preply’s discovered pricing page still points to subscriptions and corporate language training, showing that tutor-led models also package recurring revenue and B2B offers. | 中 | SI016 |
| CI023 | ELSA’s discovered pricing page points to Pro Memberships, Business, and Schools, highlighting that specialized speaking apps can monetize across consumer and institutional channels. | 中 | SI015 |
| CI024 | Busuu’s premium page title and Babbel’s pricing/business links confirm that broad-course competitors use explicit paid upsells rather than purely free distribution. | 中 | SI013, SI014 |
| CI025 | Praktika markets roughly $8/month AI tutoring, well below Speak’s public price anchor, which pressures category willingness to pay if AI practice commoditizes. | 中 | SI025, SI002 |
| CI026 | Loora’s business-facing and app-download surface suggests another AI tutor pursuing both consumer and business monetization, further compressing differentiation. | 中 | SI020, SI026 |
| CI027 | Speak’s unit economics remain largely opaque because public sources do not reveal CAC, payback, gross margin, or retention by cohort. | 低 | |
| CI028 | A reasonable public reading is that Speak has moderate capital intensity: software, AI inference, content/localization, and app-store distribution costs, but no visible inventory or hardware capex burden. | 中 | SI001, SI002, SI003 |
| CI029 | AI inference and feedback quality likely make Speak’s delivery costs higher than static course apps, even though public sources do not quantify the margin impact. | 中 | SI001, SI008 |
| CI030 | App-store dependence implies revenue collection and discovery are mediated by platform policies and fees rather than wholly controlled by Speak. | 中 | SI002, SI003 |
| CI031 | Speak’s sales-efficiency profile on the enterprise side cannot be publically modeled because buyer count is disclosed but no seat counts, ACVs, or sales-cycle data are provided. | 中 | SI004, SI006 |
| CI032 | The company’s valuation has risen faster than its public operating disclosure, creating a gap between financing confidence and financial transparency. | 中 | SI005, SI006, SI007, SI008 |
| CI033 | There is no reviewed public sign of debt, project-finance obligations, or acute capital distress, but there is also no evidence to prove strong runway. | 中 | SI005, SI006, SI009, SI010 |
| CI034 | New-language expansion broadens what Speak can sell, but no public source quantifies attach rate, ARPU by language, or whether new courses monetize at the same price. | 中 | SI001, SI005, SI006 |
| CI035 | The strongest public case for Speak’s near-term financing adequacy is simply that it raised significant capital recently and has not publicly disclosed distress signals. | 中 | SI005, SI006, SI010 |
| CI036 | The weakest part of the financial story is revenue quality: users, downloads, and valuations are public, but realized revenue, margins, and retention are not. | 中 | SI001, SI005, SI006 |
| CI037 | If Speak’s enterprise motion is meaningful, working-capital needs likely come from sales, onboarding, and localization rather than physical fulfillment. | 中 | SI004, SI006 |
| CI038 | Category benchmarks show consumers can choose among subscription apps, freemium bundles, live tutoring, tutor marketplaces, and free exchange communities, which caps pricing power for a mid-priced app like Speak. | 中 | SI012, SI017, SI018, SI023, SI025 |
| CE001 | Speak publicly positions itself as an AI language tutor centered on speaking out loud and instant feedback. | 高 | SE001, SE017 |
| CE002 | Public product surfaces show Speak teaching six target languages: Spanish, English, French, Italian, Japanese, and Korean. | 高 | SE001, SE017, SE019 |
| CE003 | Speak’s homepage says the app has 15M+ downloads. | 中 | SE001 |
| CE004 | Speak’s proprietary learning method is organized around Learn, Practice, and Apply phases. | 高 | SE005, SE006, SE007 |
| CE005 | Tutor Lessons use an AI voice agent that can advance, correct mistakes, answer clarifying questions, and redirect off-topic responses. | 中 | SE005 |
| CE006 | Apply-phase roleplays are open-ended conversations where learners complete objectives without one fixed correct answer. | 中 | SE005, SE003 |
| CE007 | Speak said in March 2023 that GPT-4 had already been in production for two months powering parts of AI Tutor. | 中 | SE029 |
| CE008 | Speak’s winter 2025 release added adaptive lessons, vocab side quests, unit refreshers, audio-first roleplays, and Speak Level. | 中 | SE007 |
| CE009 | Speak launched French, Japanese, Korean, and Italian for English speakers in June 2025 after an earlier Spanish release. | 高 | SE008, SE010 |
| CE010 | Before its 2024 ASR overhaul, Speak operated fragmented speech systems across iOS, Android, on-device models, and third-party recognition services. | 中 | SE002 |
| CE011 | Speak’s 2024 ASR upgrade fine-tuned Conformer-CTC on many thousands of hours of heavily accented learner speech. | 中 | SE002 |
| CE012 | Speak’s 2024 speech stack used Nvidia Riva and Triton on Kubernetes, Google Cloud, gRPC between services, and websockets for client-server streaming. | 中 | SE002 |
| CE013 | Speak reported a greater than 60% word-error-rate reduction versus its pre-trained Conformer baseline on internal learner data. | 中 | SE002 |
| CE014 | Speak reported a 45% WER improvement versus its earlier fine-tuned on-device Android model. | 中 | SE002 |
| CE015 | Speak said first-word feedback latency averaged about 1.6 seconds after the 2024 ASR upgrade, about 20% faster than its prior third-party service. | 中 | SE002 |
| CE016 | Matching v2 replaces bag-of-words matching with a combined ASR-plus-phonetic pipeline and forced alignment. | 中 | SE004 |
| CE017 | Speak’s matching pipeline uses a customized wav2vec2 family model for phonetic streaming inference and updates transcripts every 200–300 milliseconds while a user speaks. | 中 | SE004 |
| CE018 | Speak reported that Matching v2 reduced false negatives by about 40% without increasing false positives on internal labeled data. | 中 | SE004 |
| CE019 | Speak’s 2026 voice-agent platform uses WebRTC via LiveKit Cloud between mobile clients and its backend. | 中 | SE005 |
| CE020 | Speak’s voice-agent servers are built on LiveKit Agents and call external ASR, LLM, TTS, and speech-to-speech providers alongside Speak backend services. | 中 | SE005 |
| CE021 | Speak says its Kubernetes clusters run across multiple regions and route learners to geographically close LiveKit edges and Speak clusters. | 中 | SE005 |
| CE022 | Speak uses both cascade (ASR→LLM→TTS) and speech-to-speech pipelines rather than standardizing on one architecture. | 中 | SE005 |
| CE023 | Speak says cascade fits roleplays and free-form conversations, while speech-to-speech fits pronunciation feedback and tutor lessons where tone or accent matters. | 中 | SE005 |
| CE024 | Speak evaluates and mixes multiple TTS providers by language pair, code-switching quality, latency, and custom-voice fit because no single provider works best everywhere. | 中 | SE005 |
| CE025 | Speak tracks end-to-end latency, ASR time-to-final-transcript, TTS time-to-first-byte, and provider performance by region, language, and tail latency percentiles. | 中 | SE005 |
| CE026 | Speak says it automatically shifts traffic to backup providers when latency or error thresholds are exceeded in a region-language pair. | 中 | SE005 |
| CE027 | Live Roleplays combine OpenAI Realtime API with Speak’s proprietary learning engine, proficiency graph, objectives, and hints. | 中 | SE003 |
| CE028 | Speak acknowledged in October 2024 that new speech-to-speech models still lag text models on instruction following and nuanced pronunciation coaching. | 中 | SE003 |
| CE029 | Speak says all lessons are written by learning designers and then human-reviewed even when AI is used to speed up workflow. | 中 | SE011 |
| CE030 | Speak says a single four-unit course can include 50+ lessons and takes weeks of writing, revising, filming, testing, and polishing. | 中 | SE006 |
| CE031 | Speak publicly supports iOS and Android mobile devices but not desktop PCs. | 中 | SE012 |
| CE032 | As of March 2026, Speak lists minimum support at app version 4.35.0, iOS 16, and Android 8. | 中 | SE012 |
| CE033 | Speak exposes both an in-lesson issue-report flag and email-based support escalation workflows. | 高 | SE014, SE015 |
| CE034 | Google Play says Speak may share app activity and device IDs with third parties, may collect personal and financial information, encrypts data in transit, and supports data deletion requests. | 中 | SE019 |
| CE035 | Apple’s listing rates Speak 13+ and links users to a privacy policy and terms page hosted on usespeak.com/speak.com domains. | 中 | SE017 |
| CE036 | Speak’s public privacy and terms URLs returned only a JavaScript shell in text-only review, limiting direct inspection of policy details. | 中 | SE030, SE031 |
| CE037 | OpenAI’s April 2025 interview says Speak uses OpenAI models across audio and text and that Connor Zwick viewed the Realtime API plus multimodal audio as a key breakthrough for the product. | 中 | SE023 |
| CE038 | TechCrunch reported in December 2024 that Speak had more than 10 million downloads and over 200 Speak for Business customers. | 中 | SE024 |
| CE039 | SiliconANGLE reported that Live Roleplays rolled out in late 2024 and that Speak also supports business-oriented conversations with suppliers and customers. | 中 | SE025 |
| CE040 | Apple’s App Store page shows Speak at 4.8 stars from 44K ratings. | 中 | SE017 |
| CE041 | Google Play shows Speak at 4.7 stars from roughly 112K reviews and 10M+ downloads as of the May 2026 listing. | 中 | SE019 |
| CE042 | Recent Apple reviews consistently praise Speak’s immediate speaking practice, replay/pronunciation comparison, and AI conversations, but also ask for more languages and richer rewards. | 中 | SE018 |
| CE043 | JustUseApp surfaces public complaints about laggy voice recognition, refund friction, support delays, and unfinished lesson availability despite strong headline ratings. | 中 | SE020 |
| CE044 | LanguaTalk’s 2026 review says Speak’s voices and speech recognition are strong, but feedback depth, lesson variety at higher levels, and pricing clarity lag serious-learner expectations. | 中 | SE021 |
| CE045 | AppsHunter shows the iOS app updated April 30, 2026 at version 4.46.0 with iOS 16+ compatibility and 16 interface languages. | 中 | SE026 |
| CE046 | OpenAI’s developer community contains active 2025 implementation guidance for realtime voice agents using Twilio, FastAPI, and OpenAI’s Realtime API. | 中 | SE027 |
| CE047 | The public GitHub repository openai/openai-realtime-agents showed about 6.8k stars and 1.1k forks at review time, indicating an active ecosystem around a key Speak dependency. | 中 | SE028 |
| CE048 | Speak says its English-learning product now supports 15 different native-language entry points. | 中 | SE010 |
| CE049 | Speak said in November 2024 that Google Play named it Best App of 2024 in Hong Kong, Korea, and Taiwan. | 中 | SE009 |
| CE050 | Speak’s help center says more intermediate courses and broader non-English access are still in development. | 中 | SE010, SE011 |
| CU001 | Speak's homepage positions the product as an AI language tutor focused on speaking out loud with instant feedback. | 中 | SU001 |
| CU002 | Speak's homepage lists French, Spanish, English, Korean, Italian, and Japanese as supported learning languages. | 中 | SU001 |
| CU003 | Speak's homepage advertises a 4.8 rating. | 中 | SU001 |
| CU004 | Speak's homepage advertises 15M+ downloads. | 中 | SU001 |
| CU005 | Speak for Business says 200+ brands rely on the product. | 中 | SU002, SU005 |
| CU006 | Speak for Business frames employers as the buyer/payer and employees as the primary users of English-learning content. | 中 | SU002 |
| CU007 | Speak for Business says learners speak hundreds of sentences per week. | 中 | SU002 |
| CU008 | Speak's official review page is explicitly limited to real 5-star App Store reviews from U.S. learners. | 中 | SU003 |
| CU009 | Speak said in June 2024 that it had more than 10 million learners in 40+ countries. | 中 | SU004, SU007 |
| CU010 | Speak said in June 2024 that learners had more than doubled year over year for the previous five years. | 中 | SU004, SU007 |
| CU011 | Speak said learners speak 1,000 times on average in their first week. | 中 | SU004 |
| CU012 | Speak said nearly 6% of Korea's population was learning English with the app in 2024. | 中 | SU004, SU008 |
| CU013 | Speak said users had already spoken more than one billion sentences in 2024. | 中 | SU005 |
| CU014 | Speak said it created 25 million personalized lessons in 2024. | 中 | SU005 |
| CU015 | Speak said Speak for Business had more than 200 customers across industries in December 2024. | 中 | SU005, SU012 |
| CU016 | Speak said enterprise deployments were seeing an 85% employee adoption rate in December 2024. | 中 | SU005 |
| CU017 | Speak Wrapped 2025 says learners spoke 3.74 billion lines in 2025, up 111% year over year. | 中 | SU006 |
| CU018 | Speak Wrapped 2025 says learners spent 19.6 million hours practicing in app in 2025, up 85% year over year. | 中 | SU006 |
| CU019 | Speak Wrapped 2025 says learners started 231 million lessons in 2025, up 108% year over year. | 中 | SU006 |
| CU020 | Speak Wrapped 2025 says learners created 80.3 million personalized lessons in 2025, up 154% year over year. | 中 | SU006 |
| CU021 | TechCrunch reported in August 2023 that Speak was live in around 20 countries. | 中 | SU008 |
| CU022 | TechCrunch reported in August 2023 that Speak had well over 100,000 subscribers in South Korea. | 中 | SU008 |
| CU023 | Apple's App Store listing shows Speak at 4.8 out of 5 from 44K ratings. | 高 | SU009, SU010 |
| CU024 | Google Play shows Speak at 4.7 stars from 112K reviews and 10M+ downloads. | 中 | SU011 |
| CU025 | The Google Play listing shows the Android app was updated on 2026-05-01. | 中 | SU011 |
| CU026 | The iOS App Store listing shows monthly Premium pricing at $17.99, annual Premium at $83.99, and annual Premium Plus at $164.99. | 中 | SU018 |
| CU027 | Forbes reported in November 2025 that about 15 million people had downloaded Speak and the company had surpassed $100 million in annualized revenue, largely from consumers. | 低 | SU013 |
| CU028 | Forbes reported that Speak began pushing into enterprise in 2024 after some consumers asked employers to cover subscriptions. | 低 | SU013 |
| CU029 | Forbes reported that roughly 500 companies, including KPMG and HD Hyundai, offered Speak subscriptions to employees primarily in South Korea by late 2025. | 低 | SU013 |
| CU030 | KPMG describes itself as a global organization of independent professional services firms providing audit, tax, and advisory services. | 中 | SU017 |
| CU031 | The App Store and Speak's official review page both preserve j herronov's review saying months of French use improved comprehension of French media and made free-talk and bookmark features valuable. | 中 | SU019, SU026 |
| CU032 | Google Play and Speak's official review page both preserve Dan S's review saying more than six months of Spanish use beat prior tools because feedback was detailed and instantaneous. | 中 | SU021, SU024 |
| CU033 | Google Play and Speak's official review page both preserve Rosalyn Mulder's review saying tutor-like feedback and streaks made Speak preferable to Duolingo and Mango. | 中 | SU022, SU025 |
| CU034 | The App Store reviews page shows Chuck_Ellis saying eight days of use had him forming Spanish sentences with confidence. | 中 | SU010 |
| CU035 | The App Store reviews page shows dollargills saying Speak helps with speaking and listening rather than only vocabulary memorization. | 中 | SU010 |
| CU036 | The App Store listing includes brombres' review praising focused speaking practice while calling the UI and HFM auto-advance behavior unintuitive. | 中 | SU020 |
| CU037 | The Google Play listing includes Danielle Chavez's April 2026 review saying speech recognition misses words mid-sentence and female verb forms can be mishandled. | 中 | SU023 |
| CU038 | MWM's editorial app page reports 10M+ downloads, a 4.8 out of 5 user rating, and 354.9K total ratings across locales. | 低 | SU016 |
| CU039 | LanguaTalk rated Speak 3 out of 5 in February 2026 and said the app works best for beginners while feedback depth and lesson variety remain weaker for serious learners. | 中 | SU015 |
| CU040 | LanguaTalk said Speak's speech recognition can be overly lenient and may give learners a false sense of mastery. | 中 | SU015 |
| CU041 | JustUseApp surfaces complaints about refund friction, language availability, voice recognition, and missing Spanish-content access. | 低 | SU014 |
| CU042 | App-store listings on Apple and Google both use auto-renewing subscriptions, indicating Speak's public monetization flow is optimized for self-serve consumer conversion. | 中 | SU009, SU011 |
| CU043 | Because Speak's official review page only republishes 5-star App Store reviews, it cannot by itself prove balanced customer satisfaction or retention. | 中 | SU003 |
| CU044 | Speak's public customer proof is much stronger for consumer learners and aggregate enterprise counts than for named enterprise case studies. | 中 | SU002, SU003, SU013 |
| CU045 | No reviewed public source disclosed NRR, GRR, churn percentages, or cohort-retention tables for Speak. | 中 | SU001, SU002, SU005, SU013, SU015 |
| CU046 | No reviewed public source disclosed contract lengths, renewal rates, or seat-expansion data for Speak for Business. | 中 | SU002, SU005, SU013 |
| CU047 | No reviewed public source disclosed top-customer concentration, regional revenue mix, or partner concentration for Speak. | 中 | SU002, SU005, SU013, SU016 |
| CU048 | Public named-customer proof is sampled rather than exhaustive: official and press sources cite enterprise counts but do not provide a public roster or case-study library for most employer logos. | 中 | SU002, SU005, SU013 |
| CU049 | Dataconomy reported that Speak users spend about 10 to 20 minutes per day in the app. | 低 | SU012 |
| CU050 | Dataconomy reported that Speak for Business served over 200 corporate customers in December 2024. | 中 | SU012 |
| CR001 | Apple’s App Store listing shows Speak with a 4.8 rating from 44K ratings and a 13+ age rating. | 中 | SR004 |
| CR002 | Google Play shows Speak with 10M+ downloads, 112K reviews, and a 4.7 star rating. | 中 | SR013 |
| CR003 | Google Play’s data-safety disclosure says Speak may share app activity and device or other IDs with third parties. | 中 | SR013 |
| CR004 | Google Play’s data-safety disclosure says Speak may collect personal info, financial info, and four other data categories. | 中 | SR013 |
| CR005 | Google Play says Speak encrypts data in transit and lets users request deletion. | 中 | SR013 |
| CR006 | Speak’s Google Play listing says memberships auto-renew unless cancelled at least 24 hours before renewal. | 中 | SR013 |
| CR007 | Speak’s Apple App Store listing says memberships auto-renew unless cancelled at least 24 hours before renewal. | 中 | SR004 |
| CR008 | Speak’s cancellation article says cancelling a subscription prevents the next renewal but does not generate a refund for the current period. | 中 | SR010 |
| CR009 | Speak routes cancellation differently for Apple App Store, Google Play, and direct website purchases. | 中 | SR010 |
| CR010 | Speak’s refund policy gives Google Play and website subscribers a refund window of seven days from purchase. | 中 | SR011 |
| CR011 | Speak’s refund policy says no refunds are available after 30 days from payment for Google Play and website purchases. | 中 | SR011 |
| CR012 | Speak’s refund policy says Apple controls App Store refunds and Speak cannot process those refunds directly. | 中 | SR011 |
| CR013 | Apple Support says users can cancel in-app subscriptions and request refunds through Apple. | 中 | SR014 |
| CR014 | Google Play tells developers to be transparent about subscription terms, billing frequency, and cancellation methods. | 中 | SR015 |
| CR015 | Google Play says subscription apps must include an easy-to-use online method to cancel the subscription. | 中 | SR015 |
| CR016 | COPPA applies when an online service is directed to children under 13 or has actual knowledge that it collects personal information from children under 13. | 中 | SR008, SR020 |
| CR017 | FTC COPPA guidance says covered services must post a privacy policy and obtain verifiable parental consent before collecting personal information from children. | 中 | SR020 |
| CR018 | CNIL says the AI Act was published in the Official Journal in July 2024 and entered into force in stages from 1 August 2024. | 中 | SR018 |
| CR019 | CNIL says the AI Act is risk-based and includes high-risk examples such as biometric systems plus specific transparency obligations for some AI systems. | 中 | SR018 |
| CR020 | IAPP’s AI Act/GDPR mapping says some high-risk AI deployments require a fundamental-rights impact assessment that complements GDPR impact assessments. | 中 | SR019 |
| CR021 | IAPP says GDPR Articles 13-14, 15, and 22 create transparency and meaningful-information duties around automated decision-making. | 中 | SR019 |
| CR022 | OpenAI says Speak uses OpenAI models across audio and text modalities for interactive speaking exercises and tutors. | 中 | SR002 |
| CR023 | Speak’s Live Roleplays feature is built with OpenAI’s Realtime API and GPT-4o speech-to-speech capabilities. | 中 | SR024, SR007 |
| CR024 | Speak says current speech-to-speech models are still weaker than text models on instruction following and nuanced language-learning tasks such as pronunciation coaching. | 中 | SR024 |
| CR025 | Speak’s ASR team says earlier third-party speech-recognition services often struggled with heavily accented learner speech. | 中 | SR001 |
| CR026 | Speak runs its fine-tuned Conformer-CTC ASR stack with Nvidia Riva and Triton in Kubernetes on Google Cloud Platform. | 中 | SR001 |
| CR027 | Speak says the revamped ASR system delivers first-word feedback in about 1.6 seconds on average. | 中 | SR001 |
| CR028 | Speak said in March 2023 that GPT-4 had already powered parts of AI Tutor in production for more than two months and over 2M lessons had used the feature. | 中 | SR025 |
| CR029 | TechCrunch reported in June 2024 that Speak had grown to over 10 million users and customers in more than 40 countries. | 中 | SR030 |
| CR030 | TechCrunch reported in June 2024 that Speak’s user base had doubled every year for the prior five years. | 中 | SR030 |
| CR031 | Speak said in August 2023 that nearly 6% of South Korea’s population had used the app. | 中 | SR026 |
| CR032 | Speak said in August 2023 that it had begun international expansion and was live in more than 20 countries. | 中 | SR026 |
| CR033 | TechCrunch reported in December 2024 that Speak raised a $78M Series C at a $1B valuation. | 中 | SR027, SR007 |
| CR034 | SiliconANGLE reported that Speak’s plan to build custom LLMs could create significant costs that the new financing helps absorb. | 中 | SR007 |
| CR035 | JustUseApp’s review page says 67.1% of the combined experience it analyzed was negative. | 低 | SR005 |
| CR036 | JustUseApp includes complaints that Speak’s voice recognition became laggy and did not allow enough time to finish some spoken sentences. | 低 | SR005 |
| CR037 | JustUseApp includes complaints about unauthorized or unexpected charges after free-trial cancellation and about support not responding for a week on refund requests. | 低 | SR005 |
| CR038 | JustUseApp includes a complaint that Speak required a Korean phone number during onboarding and another that the interface launched in a language the user did not understand. | 低 | SR005 |
| CR039 | AppsHunter summarizes negative user themes as expensive subscriptions, inaccurate voice recognition, rushed lessons, limited languages, and occasional app crashes or bugs. | 低 | SR006 |
| CR040 | AppsHunter lists reported bugs including inability to record during lessons, lesson progress not saving properly, and billing problems. | 低 | SR006 |
| CR041 | Tooliverse flags premium pricing, pronunciation misreads, battery heating during long sessions, and background-noise sensitivity among its key watch-outs for Speak. | 低 | SR022 |
| CR042 | A 2026 Google Play review says the app sometimes fails to hear everything a user says in the middle of a sentence. | 低 | SR013 |
| CR043 | Apple App Store reviews and Speak’s curated review page both show strong user praise for speaking confidence gains and immediate feedback, but repeated requests for more languages. | 中 | SR012, SR023 |
| CR044 | Apple rates Speak 13+ while Google Play rates it Everyone, creating an ambiguous minor-facing posture for a voice-first tutoring product. | 中 | SR004, SR013, SR008 |
| CR045 | Because billing and refunds are split across Apple, Google Play, and Speak’s direct website, subscription-trust failures can be operationally fragmented and harder to resolve quickly. | 中 | SR010, SR011, SR013, SR014 |
| CR046 | OpenAI price, policy, or uptime changes can affect both feature quality and gross-margin assumptions because Speak publicly relies on GPT-4, GPT-4o, and Realtime API capabilities. | 中 | SR002, SR024, SR027 |
| CR047 | Speak’s cloud dependency is concentrated in a narrow infrastructure stack because its custom ASR backend runs on Google Cloud with Nvidia GPU inference components. | 中 | SR001 |
| CR048 | Rapid growth from South Korea into 20-plus and then 40-plus countries raises localization, support, and content-operations complexity even if demand remains strong. | 中 | SR026, SR030, SR029 |
| CR049 | The public review and help-center record shows billing friction is not a one-off issue because refunds, cancellation, payment status, and free-trial questions occupy a dedicated 11-article help collection alongside multiple complaint sources. | 中 | SR003, SR005, SR010, SR011 |
| CR050 | Public review evidence supports the idea that Speak has real product value, but premium pricing makes customer expectations for recognition quality and billing fairness materially higher. | 中 | SR012, SR022, SR005 |
| CR051 | NicheMetric estimates Speak generated about $10.0M of iOS revenue and more than 2.0M iOS downloads in the last 30 days, but the methodology is not transparent enough for high-confidence underwriting. | 低 | SR028 |
| CR052 | SiliconANGLE says Speak now offers an enterprise edition, Speak for Business, with business-conversation features, increasing the need for support and contract maturity beyond the consumer app. | 中 | SR007 |
| CR053 | Y Combinator’s company page lists Speak as an active San Francisco company with multiple open jobs, which is directionally consistent with ongoing staffing needs rather than a fully stabilized operating model. | 低 | SR029 |
| CR054 | Speak’s publicly reviewable privacy and terms pages were JavaScript-gated during chapter preparation, so voice-retention, transcript-retention, and model-training specifics could not be fully verified from machine-readable text. | 低 | |
| CR055 | Public sources did not disclose Speak’s burn rate, gross margin, inference cost per lesson, customer concentration, or enterprise SLA credits. | 低 | |
| CR056 | No comprehensive cross-jurisdiction litigation or enforcement package was publicly reviewable during chapter preparation, so legal-overhang risk remains only partially closed. | 低 | |
| CV001 | Speak announced a $20M Series B-3 on June 18, 2024 at a $500M valuation. | 中 | SV001, SV003 |
| CV002 | Speak said the June 2024 financing brought total funding to $84M. | 中 | SV001, SV003 |
| CV003 | Speak said it had more than 10 million learners in 40+ countries by June 2024. | 中 | SV001, SV003 |
| CV004 | Speak said its learner base had more than doubled year over year for five straight years by June 2024. | 中 | SV001, SV003 |
| CV005 | TechCrunch reported Speak charged $20 per month or $99 per year in mid-2024. | 中 | SV003 |
| CV006 | Speak announced a $78M Series C on December 10, 2024 at a $1B valuation. | 中 | SV002, SV006, SV004 |
| CV007 | Speak said the Series C brought lifetime funding to $162M. | 中 | SV002, SV004 |
| CV008 | Speak said users had already spoken more than one billion sentences with the product in 2024. | 中 | SV002 |
| CV009 | Speak said Speak for Business had 200+ customers and an 85% employee adoption rate in December 2024. | 中 | SV002, SV006 |
| CV010 | Forbes said Speak’s valuation doubled to $1B after the December 2024 $78M round. | 中 | SV004, SV002 |
| CV011 | Forbes reported roughly 15 million people had downloaded Speak by November 2025. | 中 | SV005 |
| CV012 | Forbes reported Speak had surpassed $100M in annualized revenue by November 2025. | 中 | SV005, SV007 |
| CV013 | Forbes reported about 500 companies, including KPMG and HD Hyundai, offered Speak subscriptions to employees by late 2025. | 中 | SV005 |
| CV014 | Forbes reported paid Speak access ranged from roughly $80 to $200 for consumers in late 2025. | 中 | SV005, SV009 |
| CV015 | Apple’s App Store listed Speak at 4.8 stars with 44K ratings when accessed on 2026-05-05. | 中 | SV009 |
| CV016 | Apple’s App Store listed current U.S. in-app prices including $17.99 monthly premium and $83.99 annual premium, with higher Plus tiers. | 中 | SV009 |
| CV017 | Google Play listed Speak at 10M+ downloads and 112K reviews when accessed on 2026-05-05. | 中 | SV010 |
| CV018 | Google Play showed Speak was updated on May 1, 2026. | 中 | SV010 |
| CV019 | An AppBrain snapshot from February 2025 estimated 5.8M Android lifetime downloads and about 350K recent 30-day downloads for Speak. | 中 | SV011 |
| CV020 | An AppBrain snapshot showed Speak ranked #1 top grossing in South Korea Education and #2 top grossing in Japan Education in early 2025. | 中 | SV011 |
| CV021 | NicheMetric estimated Speak generated about $10.0M in iOS revenue and more than 2.0M iOS downloads in the last 30 days. | 中 | SV008 |
| CV022 | NicheMetric surfaced critical 2026 user complaints about an immediate paywall, poor support responsiveness, and inability to switch languages. | 中 | SV008 |
| CV023 | GetLatka reported Speak reached $100M revenue in 2025, up from $15M in 2024. | 中 | SV007 |
| CV024 | GetLatka listed Speak at 253 employees. | 中 | SV007 |
| CV025 | Yahoo Finance listed Duolingo at $1.04B trailing revenue and 4.01x EV/Revenue. | 中 | SV014 |
| CV026 | Yahoo Finance listed Duolingo at 39.91% profit margin and 35.0% quarterly revenue growth. | 中 | SV014 |
| CV027 | CompaniesMarketCap listed Duolingo at a $5.15B market cap in May 2026. | 中 | SV015 |
| CV028 | SEC EDGAR showed Duolingo had filed a current 10-Q on 2026-05-05 and a 10-K for 2025. | 中 | SV013 |
| CV029 | Yahoo Finance listed Coursera at $789.84M trailing revenue and 0.42x EV/Revenue. | 中 | SV017 |
| CV030 | Yahoo Finance listed Coursera at -3.0% quarterly revenue growth and 0.48% profit margin. | 中 | SV017 |
| CV031 | CompaniesMarketCap listed Coursera at a $0.98B market cap in May 2026. | 中 | SV018 |
| CV032 | SEC EDGAR showed Coursera had a 10-Q on 2026-04-30 and recent annual-report filings on file. | 中 | SV016 |
| CV033 | Yahoo Finance listed Udemy at $773.9M trailing revenue and 0.38x EV/Revenue. | 中 | SV020 |
| CV034 | Yahoo Finance listed Udemy at 9.1% quarterly revenue growth and -8.23% profit margin. | 中 | SV020 |
| CV035 | CompaniesMarketCap listed Udemy at a $0.68B market cap in May 2026. | 中 | SV021 |
| CV036 | SEC EDGAR showed Udemy had a 2026 10-K/A and prior 10-K disclosures on file. | 中 | SV019 |
| CV037 | Yahoo Finance listed Chegg at $376.91M trailing revenue and 0.32x EV/Revenue. | 中 | SV023 |
| CV038 | Yahoo Finance listed Chegg at -49.4% quarterly revenue growth and -27.44% profit margin. | 中 | SV023 |
| CV039 | CompaniesMarketCap listed Chegg at a $0.12B market cap in May 2026. | 中 | SV024 |
| CV040 | SEC EDGAR showed Chegg had 2026 10-Qs and a 2025 10-K on file. | 中 | SV022 |
| CV041 | Grand View Research estimated the global AI tutors market at $2.11B in 2025 and $17.72B by 2033, a 30.5% CAGR. | 中 | SV027 |
| CV042 | MMR Statistics estimated the global online language learning market at $24.56B in 2025 and $63.43B by 2032, a 14.52% CAGR. | 中 | SV028 |
| CV043 | MMR Statistics said more than 60% of leading online-language platforms had integrated AI-driven adaptive learning, speech recognition, or personalized lesson pathways in 2025. | 中 | SV028 |
| CV044 | MMR Statistics said freemium pricing and free content were intensifying competition and raising customer acquisition costs, especially for earlier-stage platforms. | 中 | SV028 |
| CV045 | Fortune Business Insights sized the private tutoring market at $66.96B in 2025 and said Asia Pacific held a 60.85% share. | 中 | SV029 |
| CV046 | The FTC warned that control over key generative-AI inputs can create barriers to entry and distort competition. | 中 | SV026 |
| CV047 | The FTC warned network effects can help generative-AI leaders entrench market power and reduce entrant competitiveness. | 中 | SV026 |
| CV048 | Oliver Wyman said public markets began repricing software risk in early 2026 as agentic AI challenged seat-based pricing and durable product differentiation. | 中 | SV025 |
| CV049 | Oliver Wyman said valuation multiples are becoming more sensitive to perceived AI exposure and revenue durability. | 中 | SV025 |
| CV050 | Using Speak’s $1.0B December 2024 valuation and the later reported >$100M annualized revenue milestone implies a high-single-digit to roughly 10x revenue multiple before later balance-sheet adjustments. | 中 | SV002, SV005 |
| CV051 | Speak’s implied private multiple sits above current public-peer EV/revenue levels of 4.01x for Duolingo, 0.42x for Coursera, 0.38x for Udemy, and 0.32x for Chegg. | 中 | SV014, SV017, SV020, SV023 |
| CV052 | The public-comp spread suggests Speak’s premium valuation only holds if it sustains materially faster growth and monetization than listed edtech peers. | 中 | SV014, SV017, SV020, SV023, SV005 |
| CV053 | Public evidence supports real traction, but absent cap-table, preference-stack, retention, and segment-mix disclosure makes the last public $1B mark hard to underwrite for a new investor. | 中 | SV002, SV005, SV007 |
| CV054 | A reasonable bear case is valuation compression toward roughly $400M-$650M if growth slows and public-like multiples dominate. | 中 | SV014, SV017, SV020, SV023, SV025 |
| CV055 | A reasonable base case of roughly $800M-$1.0B requires Speak to defend current scale and convert enterprise traction without requiring multiple expansion. | 中 | SV002, SV005, SV009, SV010 |
| CV056 | A reasonable bull case above $1.1B requires revenue scaling well beyond $150M plus broader enterprise adoption and continued app-ranking strength. | 中 | SV005, SV011, SV027, SV028 |
| CV057 | Because market growth is real but AI competition and multiple compression are also real, a research-more recommendation is better supported than a buy at the last public $1B mark. | 中 | SV025, SV026, SV028, SV029, SV005 |
| CV058 | Monitorable downside triggers include weakening app-store momentum, failure to convert enterprise footprint into disclosed recurring economics, and continued sector multiple compression. | 中 | SV010, SV011, SV013, SV025, SV005 |
| 编号 | 出版方 | 标题 | 引文 |
|---|---|---|---|
| SO001 | Speak | Speak homepage | 15M+ downloads; 4.8 rating; AI language tutor focused on speaking. |
| SO002 | Speak | Speak for Business | 200+ Brands Rely on Speak for Business. |
| SO003 | Speak | Careers page | The team is based in San Francisco, Seoul, Tokyo, and Ljubljana. |
| SO004 | Apple App Store | Speak: Language Learning App Store listing | 44K Ratings; 4.8; monthly and annual auto-renewing subscriptions. |
| SO005 | Google Play | Speak: Language Learning Google Play listing | 4.7 star; 112K reviews. |
| SO006 | Speak | Series B-3 announcement | Speak raised $20M in Series B-3 financing, doubling its valuation to $500 million. |
| SO007 | Speak | Series C announcement | Speak raised $78M in Series C funding at a $1 billion valuation. |
| SO008 | Speak | New languages on Speak | Four new languages launched; more than 15 million learners around the world. |
| SO009 | TechCrunch | Language learning app Speak nets $20M, doubles valuation | Speak has grown to over 10 million users and now has customers in more than 40 countries. |
| SO010 | TechCrunch | OpenAI-backed Speak raises $78M at $1B valuation | Speak provides average usage around 10-20 minutes/day, paying $20 per month or $99 per year. |
| SO011 | Tech Funding News | OpenAI-backed Speak closes $78M at $1B valuation | Speak, a San Francisco-based language-learning startup, has closed $78 million in Series C funding. |
| SO012 | Unite.AI | Speak secures $78M Series C funding at $1B valuation | Founded in 2016 by Connor Zwick and Andrew Hsu. |
| SO013 | FilingFlow | Speakeasy Labs Form D filing | Total offering $77,699,277; first sale Nov 13, 2024; Rule 506(b). |
| SO014 | SEC Filing Data | Speakeasy Labs SEC filings list | Form D filings appear on 12/11/2024, 08/12/2024, and 10/17/2023. |
| SO015 | HolonIQ | The Complete List of Global EdTech Unicorns | Speak, Language Learning App, has joined the list in Dec 2024 at $1B valuation. |
| SO016 | GetLatka | How Speak hit $100M revenue with a 253 person team in 2025 | Speak hit $100M in revenue in November 2025 and had 253 total employees. |
| SO017 | Android Police | I ditched Duolingo for this language app, and it was a total reality check | Speak continued to praise my speaking ability even when I deliberately mispronounced phrases. |
| SO018 | JustUseApp | Speak reviews (2026) | Negative experience 67.1%; complaints include auto-renewal, lag, and speech recognition issues. |
| SO019 | Languatalk | Speak app review | Feedback is brief and lacks depth; premium tier structure is confusing and potentially expensive. |
| SO020 | Crunchbase News | AI language startup Speak hits unicorn status after raising Series C | Speak hit unicorn status after its Series C round. |
| SO021 | Inc. | This AI language learning platform is now a unicorn | Speak translated AI speech technology into a $1 billion valuation. |
| SO022 | Similarweb | Speak app overview | |
| SO023 | Sensor Tower | Speak app overview | Speak: Language Learning - Google Play Store - US - Category Rankings and Growth Metrics. |
| SO024 | Speak | Privacy policy route | |
| SO025 | Speak | Terms route | |
| SM001 | Speak | Speak homepage | The most effective way to learn a language. |
| SM002 | Speak | Series B-3 announcement | Disrupting the $100 billion+ online and in-person language learning market. |
| SM003 | Speak | Series C announcement | English learning is industry agnostic and Speak for Business had 200+ customers. |
| SM004 | Speak | New languages on Speak | Speak started by teaching English in Korea, Japan, and Taiwan. |
| SM005 | Technavio | Digital English language learning market report | Market size to increase USD 39.46 billion at a CAGR of 24.5% from 2024 to 2029. |
| SM006 | MarketsandMarkets | AI in Education Market Forecast & Size | The AI in Education market is projected to grow from USD 2.21 billion in 2024 to USD 5.82 billion by 2030. |
| SM007 | Stanford HAI | AI Index 2026 education chapter | Four out of five U.S. high school and college students now use AI for schoolwork. |
| SM008 | World Economic Forum | AI reshaping global education | AI can automate up to 20% of educator clerical tasks but raises access, privacy, and bias concerns. |
| SM009 | World Bank | Digital Progress and Trends 2025: AI Foundations | Low- and middle-income countries face steep AI adoption challenges; small AI and the four Cs matter. |
| SM010 | arXiv | Systematic Review for AI-based Language Learning Tools | AI-based language learning tools improved learner outcomes but raised privacy and teacher-preparation concerns. |
| SM011 | Preply | Global language learning statistics and trends | English is the most learned language and the English learning market is worth about $43.51B in 2025. |
| SM012 | HolonIQ | Global EdTech Unicorns list | Preply joined the EdTech unicorn list in Jan 2026. |
| SM013 | TechCrunch | OpenAI-backed Speak raises $78M at $1B valuation | For the one and a half billion people out there trying to learn English, the issue is speaking it. |
| SM014 | TechCrunch | Speak nets $20M, doubles valuation | Speak makes money by charging $20 per month, or $99 per year. |
| SM015 | Android Police | Speak review and AI limitations | Speak cannot identify basic pronunciation errors and can create a false sense of mastery. |
| SM016 | Languatalk | Speak app review | Feedback is brief and lacks depth; lesson variety becomes repetitive. |
| SM017 | SEC | Duolingo 2024 annual report | Duolingo offers courses in over 40 languages to more than 100 million monthly active users. |
| SM018 | ELSA Speak | ELSA homepage | 18M+ downloads; AI English speaking coach. |
| SM019 | Apple App Store | ELSA App Store listing | 109K ratings; yearly and monthly memberships are available. |
| SM020 | Cambly | Cambly homepage | Real conversations with native speakers, anytime, anywhere, 24/7. |
| SM021 | Busuu | Busuu App Store listing | Busuu helps you communicate with confidence from day one and connects learners with native speakers. |
| SM022 | Busuu | Busuu website snapshot | 120+ million registered users. |
| SM023 | Babbel | Babbel App Store listing | 25 million subscriptions sold. |
| SM024 | Google Play | Babbel Play listing | 50M+ downloads; 1.12M reviews. |
| SM025 | Praktika | Praktika homepage | 20M+ learners and private-tutor results without the private-tutor price. |
| SP001 | Speak | Speak homepage | 15M+ downloads; AI language tutor. |
| SP002 | Apple App Store | Speak App Store listing | 44K ratings; 4.8; free with in-app purchases. |
| SP003 | Google Play | Speak Google Play listing | 4.7 star; 112K reviews. |
| SP004 | Speak | Series C announcement | 200+ customers across various industries. |
| SP005 | TechCrunch | Speak Series B extension article | Speak makes money by charging $20 per month or $99 per year. |
| SP006 | TechCrunch | Speak Series C article | Speak for Business has over 200 customers and consumers typically pay $20 per month or $99 per year. |
| SP007 | Android Police | Speak review and AI limitations | Speak is heavily inspired by Duolingo and misses basic pronunciation errors. |
| SP008 | Languatalk | Speak app review | Speak works well for early learners but not for those seeking deeper adaptive practice. |
| SP009 | SEC | Duolingo 2024 annual report | Duolingo offers 40+ languages to 100M+ MAUs and ~9% of MAUs are paid subscribers. |
| SP010 | ELSA Speak | ELSA homepage | 18M+ downloads and 460K+ ratings. |
| SP011 | Apple App Store | ELSA App Store listing | 109K ratings; yearly and monthly memberships available. |
| SP012 | Cambly | Cambly homepage | Real conversations with native speakers, anytime, anywhere, 24/7. |
| SP013 | Busuu | Busuu website snapshot | 120+ million registered Busuu users. |
| SP014 | Apple App Store | Busuu App Store listing | 98K ratings; community feedback from native speakers. |
| SP015 | Google Play | Busuu Play listing | 50M+ downloads; 1.13M reviews. |
| SP016 | Apple App Store | Babbel App Store listing | 25 million subscriptions sold. |
| SP017 | Google Play | Babbel Play listing | 50M+ downloads; 1.12M reviews. |
| SP018 | Praktika | Praktika homepage | 20M+ learners and ~$8/month versus a ~$400/month private tutor. |
| SP019 | Loora | Loora homepage | Loora is an always-available AI English tutor focused on real-time feedback. |
| SP020 | MarketsandMarkets | AI in Education Market | Duolingo and ELSA Speak are named among the AI-in-education market participants. |
| SP021 | HolonIQ | Global EdTech Unicorns list | Preply joined the EdTech unicorn list in Jan 2026 at a $1.2B valuation. |
| SP022 | Preply | Global language learning report | English is the most learned language because of business and education demand. |
| SP023 | Speak | Speak for Business page | 200+ brands rely on Speak for Business. |
| SP024 | JustUseApp | Speak reviews aggregation | 67.1% negative experience according to review aggregation. |
| SP025 | Duolingo | Duolingo homepage | The free, fun, and effective way to learn a language. |
| SP026 | Duolingo | Duolingo Super page | Super Duolingo is the premium upsell path from the free product. |
| SP027 | Cambly | Cambly pricing page | Pricing page shows private lessons and Pro plans from US$8.12 per lesson on annual terms. |
| SP028 | Apple App Store | Cambly App Store listing | Cambly – Learn English App Store listing. |
| SP029 | Preply | Preply homepage | Preply is an online language tutoring marketplace. |
| SP030 | italki | italki homepage | italki is a language-learning marketplace with certificated tutors. |
| SP031 | HelloTalk | HelloTalk homepage | HelloTalk is a language exchange and learning platform. |
| SP032 | Google Play | Duolingo Play listing | Duolingo: Language Lessons - Apps on Google Play. |
| SP033 | Speak | New languages post | French, Japanese, Korean, and Italian launched after Spanish. |
| SI001 | Speak | Speak homepage | 15M+ downloads; AI language tutor. |
| SI002 | Apple App Store | Speak App Store listing | 44K ratings; monthly and annual auto-renewing subscriptions. |
| SI003 | Google Play | Speak Google Play listing | 4.7 star; 112K reviews. |
| SI004 | Speak | Speak for Business page | 200+ brands rely on Speak for Business. |
| SI005 | Speak | Series B-3 announcement | Speak raised $20M in Series B-3 financing, doubling its valuation to $500 million. |
| SI006 | Speak | Series C announcement | Speak raised $78M in Series C funding at a $1 billion valuation. |
| SI007 | TechCrunch | Speak Series B extension article | Speak makes money by charging $20 per month or $99 per year. |
| SI008 | TechCrunch | Speak Series C article | Speak users spend roughly 10-20 minutes per day and pay $20 per month or $99 per year. |
| SI009 | FilingFlow | Speakeasy Labs Form D filing | Total offering $77,699,277; first sale Nov 13, 2024; Rule 506(b). |
| SI010 | SECFilingData | Speakeasy Labs SEC filings list | Form D filings appear on 12/11/2024, 08/12/2024, and 10/17/2023. |
| SI011 | SEC | Duolingo 2024 annual report | Duolingo offers 40+ languages to 100M+ MAUs and about 9% of MAUs are paid subscribers. |
| SI012 | Cambly | Cambly pricing page | Private and Pro tutoring plans run from US$8.12 per lesson on annual terms. |
| SI013 | Busuu | Busuu premium page | Premium - Busuu. |
| SI014 | Apple App Store | Babbel App Store listing | 25 million subscriptions sold. |
| SI015 | ELSA Speak | ELSA pricing page | The fetched page exposes Pro Memberships, ELSA for Business, and ELSA for Schools. |
| SI016 | Preply | Preply pricing page | The fetched page points users to Preply Subscription and Corporate language training. |
| SI017 | italki | italki teachers page | 4339 English tutors available; visible trial pricing starts at USD 5.00. |
| SI018 | HelloTalk | HelloTalk VIP page | HelloTalk says it helps users learn a language for free and has 70M+ registered users across 260+ languages. |
| SI019 | Praktika | Praktika pricing page | The fetched page shows app-download CTAs and a For business link even though the pricing URL 404s. |
| SI020 | Loora | Loora pricing page | The fetched page shows Loora for Business plus app download links even though the pricing URL 404s. |
| SI021 | SEC | Duolingo Q1 2025 10-Q viewer | SEC XBRL viewer for Duolingo quarter ended March 31, 2025. |
| SI022 | ELSA Speak | ELSA homepage | 18M+ downloads and 460K+ ratings. |
| SI023 | Duolingo | Duolingo homepage | The free, fun, and effective way to learn a language. |
| SI024 | Apple App Store | Speak App Store reviews aggregation | 67.1% negative experience according to review aggregation. |
| SI025 | Praktika | Praktika homepage | 20M+ learners and roughly $8/month versus a private tutor. |
| SI026 | Loora | Loora homepage | Loora is an always-available AI English tutor. |
| SE001 | Speak | Speak - The language learning app that gets you speaking | Talk out loud, get instant feedback, and become fluent with the world’s most advanced AI language tutor. |
| SE002 | Speak | Leveling up our core speech recognition systems at Speak | This fine-tuned model dramatically outperforms the pretrained model with a >60% reduction in word error rate for our learners and task type. |
| SE003 | Speak | Live Roleplays powered by OpenAI Realtime API | Today, we’re announcing Live Roleplays, a new Speak experience that combines Realtime API with Speak’s learning engine to enable immersive, life-like speaking practice in a variety of roleplay scenarios. |
| SE004 | Speak | Designing a High-Accuracy Speech Matching Pipeline with ASR and Phonetic Models | By introducing a phonetic model alongside the ASR model, we reduced false negatives by approximately 40% without making our algorithm more lenient. |
| SE005 | Speak | Building Speak's Voice Agent Platform | Audio transport: WebRTC via LiveKit. |
| SE006 | Speak | How Speak reinvents language learning | The Speak Method is our proprietary learning method built around three phases: Learn, Practice, and Apply. |
| SE007 | Speak | Your most personalized Speak yet: What’s new in our winter release | We’re launching new features to deepen your learning experience with Speak, all designed to help you speak more, learn faster, and stay engaged every step of the way. |
| SE008 | Speak | New languages on Speak, just in time for the summer | Today we officially launch four new languages for English speakers to accompany our recent Spanish release: French, Japanese, Korean, and Italian. |
| SE009 | Speak | Speak named Google Play’s “Best App of 2024” in Hong Kong, Korea and Taiwan | Speak named Google Play’s Best App of 2024 in Hong Kong, Korea and Taiwan. |
| SE010 | Speak Help Center | What Languages Can I Learn with Speak? | Speak currently offers English learning courses for speakers of 15 native languages. |
| SE011 | Speak Help Center | How Does Speak Curate Its Content and Curriculum? | All of our lessons are written by learning designers—a team of educators, linguists, and translators. |
| SE012 | Speak Help Center | What devices and operating systems does the Speak app support? | The Speak app is available on both iOS and Android devices ... Speak is not available on desktop (PC). |
| SE013 | Speak Help Center | Voice recognition isn’t working | If Speak isn’t picking up your voice, please try the steps below. |
| SE014 | Speak Help Center | How can I report an issue in the app? | You can report an issue directly from the screen where the problem occurs. |
| SE015 | Speak Help Center | How can I contact Speak Support? | You can contact the Speak Support team by email only. |
| SE016 | Speak Help Center | Troubleshooting Guides | Common troubleshooting methods for when unexpected issues. |
| SE017 | Apple App Store | Speak: Language Learning App - App Store | Powered by cutting-edge AI technology, Speak ensures that you gain fluency by engaging in real-life conversations and receiving instant feedback. |
| SE018 | Apple App Store | Speak: Language Learning - Ratings & Reviews - App Store | I love that the lessons give you the phrase, have you repeat it and then quiz you. |
| SE019 | Google Play | Speak: Language Learning - Apps on Google Play | This app may share these data types with third parties: App activity and Device or other IDs. |
| SE020 | JustUseApp | Speak Reviews (2026) | Check if app is safe or legit | Overall Customer Experience: Negative experience 67.1% ... The voice recognition became laggy after this update. |
| SE021 | LanguaTalk | Speak App Review: Is It Worth It in 2026? | Speak’s audio quality is strong ... but the app’s feedback and customization features are unlikely to satisfy serious learners. |
| SE022 | Y Combinator | Speak: A superhuman, AI-powered language tutor in your pocket | Applied ML Engineer, Speech. |
| SE023 | OpenAI | Speak is personalizing language learning with AI | That’s easy—OpenAI’s real-time API and multimodality for audio. |
| SE024 | TechCrunch | OpenAI-backed Speak raises $78M at $1B valuation to help users learn languages by talking out loud | Speak has built a platform to teach languages by focusing on how native speakers learn: Using AI, the startup generates audio conversations and listens to users’ responses. |
| SE025 | SiliconANGLE | OpenAI backs $78M round for AI language learning startup Speak | One of the latest additions to Speak’s feature set, Live Roleplays, rolled out a few weeks ago. |
| SE026 | AppsHunter | Speak: Language Learning App - AI Speaking Practice | Updated April 30, 2026. Version 4.46.0. Compatibility: iOS 16.0+. |
| SE027 | OpenAI Developer Community | Voice Agent using Realtime API | The agent-builder package provides a streamlined way to create real-time voice agents powered by OpenAI’s Realtime API. |
| SE028 | GitHub | openai/openai-realtime-agents | This is a simple demonstration of more advanced, agentic patterns built on top of the Realtime API. |
| SE029 | Speak | Speak Shares Details of AI Tutor, Built on Top of OpenAI’s GPT-4 | Speak has used GPT-4 in production to power parts of its AI Tutor feature. |
| SE030 | Speak | Speak privacy policy landing page | Speak - The languages learning app that gets you speaking. |
| SE031 | Speak | Speak terms landing page | Speak - The languages learning app that gets you speaking. |
| SU001 | Speak | Speak homepage | 15M+ Downloads |
| SU002 | Speak | Speak for Business | Enterprise Language learning | 200+ Brands Rely on Speak for Business |
| SU003 | Speak | Why Learners Love Speak: Real App Store Feedback | This page contains a complete, unedited collection of real 5-star App Store reviews from Speak learners in the United States. |
| SU004 | Speak | Speak Hits $500M Valuation, Expands Rapidly Across Markets | Speak now has more than 10 million learners in 40+ countries, with learners more than doubling year-over-year for the last five years. |
| SU005 | Speak | A new milestone as we bring language learning to all: Raising $78M Series C at a $1B valuation | Our momentum is clear with more than 200+ customers across various industries, and an 85 percent adoption rate among employees. |
| SU006 | Speak | A year in conversation | Speak Wrapped 2025 | In 2025, learners spoke 3.74 billion lines on Speak, an increase of 111 percent from 2024. |
| SU007 | TechCrunch | Language learning app Speak nets $20M, doubles valuation | Its user base has doubled every year for the last five years, and Speak now has customers in more than 40 countries. |
| SU008 | TechCrunch | OpenAI-backed language learning app Speak raises $16M to expand to the US | Speak has managed to hold its own despite the competition, becoming one of the top-downloaded education apps in South Korea, where it first launched, with well over 100,000 subscribers. |
| SU009 | Apple App Store | Speak: Language Learning App - App Store listing | 44K Ratings |
| SU010 | Apple App Store | Speak: Language Learning - Ratings & Reviews | 4.8 out of 5 |
| SU011 | Google Play | Speak: Language Learning - Apps on Google Play | 4.7star 112K reviews 10M+ Downloads |
| SU012 | Dataconomy | AI-driven language learning startup Speak raises $78M at a $1B valuation | An enterprise tier, Speak for Business, currently serves over 200 corporate customers. |
| SU013 | Forbes | How AI Language Learning App Speak Is Taking On Duolingo | Now, some 500 companies including KPMG and HD Hyundai offer Speak subscriptions to employees primarily in South Korea. |
| SU014 | JustUseApp | Speak Reviews (2026) | Check if app is safe or legit | Paid for a subscription for Spanish and now only half of the Spanish lessons are available. |
| SU015 | LanguaTalk | Speak App Review: Is It Worth It in 2026? | Feedback is one of Speak's weakest points. |
| SU016 | MWM | Speak: Language Learning - Education App | Downloads 10M+; User Rating 4.8/5; Total Ratings 354.9K |
| SU017 | KPMG | About KPMG | KPMG is a global organization of independent professional services firms providing Audit, Tax, and Advisory services. |
| SU018 | Apple App Store | Speak pricing snapshot | Monthly Premium $17.99; Annual Premium $83.99; Annual Premium Plus $164.99 |
| SU019 | Apple App Store | j herronov App Store review | This app has improved that aspect so much for me. I am not yet fluent, but am able to pick out words and sayings on French videos and French hockey broadcasts now. |
| SU020 | Apple App Store | brombres App Store review | One small gripe: the UI doesn't feel as intuitive as it could. |
| SU021 | Google Play | Dan S Google Play review | The AI is phenominal and language recognition is dead-on. |
| SU022 | Google Play | Rosalyn Mulder Google Play review | They also have a daily streak that me, personally, I enjoy because I feel motivated to get that streak as high as possible. |
| SU023 | Google Play | Danielle Chavez Google Play review | I do run into an issue where it does not hear everything I say particularly in the middle of a sentence. |
| SU024 | Speak | Dan S review excerpt on official Speak review page | I've been using this app for more than six months now. It's simply the best Spanish language learning app that exists. |
| SU025 | Speak | Rosalyn Mulder review excerpt on official Speak review page | I absolutely love this app. There have been so many other apps that I've tried like Duolingo and Mango. They help me learn sure, but I feel this one is the best so far. |
| SU026 | Speak | j herronov review excerpt on official Speak review page | This has become by far my favorite app for a number of reasons. |
| SR001 | Speak | Leveling up our core speech recognition systems at Speak | We deployed Riva and Triton in our existing Kubernetes cluster... Our backend is deployed with Kubernetes on Google Cloud Platform. |
| SR002 | OpenAI | Speak is personalizing language learning with AI | Speak leverages OpenAI models to power its language learning curriculum across modalities such as audio and text. |
| SR003 | Speak Help Center | Subscription/Billing | Subscription/Billing ... 11 articles. |
| SR004 | Apple App Store | Speak: Language Learning App - App Store | 44K Ratings ... Age Rating 13+ ... Speak offers both monthly and annual auto-renewing subscriptions. |
| SR005 | JustUseApp | Speak Reviews (2026) | Check if app is safe or legit | Negative experience 67.1% ... The voice recognition became laggy after this update. |
| SR006 | AppsHunter | Speak: Language Learning App - AI Speaking Practice | Negative things ... Expensive subscription cost ... Voice recognition can be inaccurate ... Lesson progress not saving properly. |
| SR007 | SiliconANGLE | OpenAI backs $78M round for AI language learning startup Speak | Building such models can incur significant costs. The $78 million funding round announced today could make it easier for the company to balance those expenses with growth investments. |
| SR008 | Federal Trade Commission | Children's Online Privacy Protection Rule ("COPPA") | COPPA imposes certain requirements on operators ... directed to children under 13 years of age. |
| SR009 | European Data Protection Board | AI Privacy Risks & Mitigations Large Language Models (LLMs) | AI Privacy Risks & Mitigations Large Language Models (LLMs). |
| SR010 | Speak Help Center | How can I cancel the subscription? | If you cancel your subscription, you can continue using your subscription until the current subscription period ends, but you will not receive a refund. |
| SR011 | Speak Help Center | Refund Policy | Full refund available within 7 days of purchase ... After 30 days from the payment date: No refunds are available. |
| SR012 | Apple App Store | Speak: Language Learning - Ratings & Reviews | 4.8 out of 5 ... 44K Ratings. |
| SR013 | Google Play | Speak: Language Learning - Apps on Google Play | 10M+ Downloads ... This app may share these data types with third parties ... Data is encrypted in transit. |
| SR014 | Apple Support | Subscriptions and Billing - Official Apple Support | You can cancel a subscription from Apple ... App Store and iTunes Store purchases may be eligible for a refund. |
| SR015 | Google Play Console Help | Create and manage subscriptions | You must be transparent with users about your offer terms ... and how a user can manage or cancel their subscription. |
| SR016 | EUR-Lex | Regulation (EU) 2016/679 (GDPR) | Regulation - 2016/679 - EN - gdpr - EUR-Lex. |
| SR017 | European Data Protection Board | Artificial intelligence | Artificial intelligence | European Data Protection Board. |
| SR018 | CNIL | Entry into force of the European AI Regulation: the first questions and answers from the CNIL | The European AI Act has just been published ... and will gradually come into force as of 1 August 2024. |
| SR019 | IAPP | EU AI Act: Mapping the Interplays with the GDPR | The AI Act and the GDPR ... map interplays between the AI Act and the GDPR. |
| SR020 | Federal Trade Commission | Children’s Online Privacy Protection Rule: A Six-Step Compliance Plan for Your Business | Before collecting, using or disclosing personal information from a child, you must get their parent’s verifiable consent. |
| SR021 | Android Developers | Google Play Policies | Google Play Policies | Android Developers. |
| SR022 | Tooliverse | Speak Review 2026 - AI Language Learning | Premium pricing and battery consumption during long sessions require consideration. |
| SR023 | Speak | Why Learners Love Speak: Real App Store Feedback | This page contains a complete, unedited collection of real 5-star App Store reviews from Speak learners in the United States. |
| SR024 | Speak | Live Roleplays powered by OpenAI Realtime API | These new speech-to-speech models aren’t as good as text models on instruction following, and they’re not great yet at more nuanced language learning specific tasks. |
| SR025 | Speak | Speak Shares Details of AI Tutor, Built on Top of OpenAI’s GPT-4 | Speak has used GPT-4 in production to power parts of its AI Tutor feature. |
| SR026 | Speak | OpenAI Startup Fund-Backed Speak Announces $16m Series B-2 Financing & Rapid International Expansion | Nearly 6% of the population has turned to Speak ... now live in more than 20 countries. |
| SR027 | TechCrunch | OpenAI-backed Speak raises $78M at $1B valuation to help users learn languages by talking out loud | Speak is using the company’s technology to power its platform. |
| SR028 | NicheMetric | Speak: Language Learning - Revenue, Downloads & Market Analysis | Revenue Last 30 days $10.0M ... Downloads Last 30 days > 2.0M. |
| SR029 | Y Combinator | Speak: A superhuman, AI-powered language tutor in your pocket | Active ... San Francisco ... Jobs 10. |
| SR030 | TechCrunch | Language learning app Speak nets $20M, doubles valuation | Speak has grown to over 10 million users ... customers in more than 40 countries. |
| SV001 | Speak | Speak Hits $500M Valuation, Expands Rapidly Across Markets | |
| SV002 | Speak | A new milestone as we bring language learning to all: Raising $78M Series C at a $1B valuation | |
| SV003 | TechCrunch | Language learning app Speak nets $20M, doubles valuation | |
| SV004 | Forbes via Internet Archive | Speak | Company Overview & News | |
| SV005 | Forbes via Internet Archive | This Startup Is Racing Duolingo To Replace Human Language Tutors With AI | |
| SV006 | AIbase | AI Language Learning Platform Speak Raises $78 Million, Valuation Exceeds $1 Billion | |
| SV007 | GetLatka | How Speak hit $100M revenue with a 253 person team in 2025. | |
| SV008 | NicheMetric | Speak: Language Learning - Revenue, Downloads & Market Analysis | |
| SV009 | Apple App Store | Speak: Language Learning App - App Store | |
| SV010 | Google Play | Speak: Language Learning - Apps on Google Play | |
| SV011 | AppBrain via Internet Archive | Speak - Language Learning for Android - Free App Download | |
| SV012 | Apptopia | About: Speak: Language Learning (Google Play version) | |
| SV013 | SEC EDGAR | EDGAR Company Search Results - Duolingo, Inc. | |
| SV014 | Yahoo Finance | Duolingo, Inc. (DUOL) Valuation Measures & Financial Statistics | |
| SV015 | CompaniesMarketCap | Duolingo (DUOL) - Market capitalization | |
| SV016 | SEC EDGAR | EDGAR Company Search Results - Coursera, Inc. | |
| SV017 | Yahoo Finance | Coursera, Inc. (COUR) Valuation Measures & Financial Statistics | |
| SV018 | CompaniesMarketCap | Coursera (COUR) - Market capitalization | |
| SV019 | SEC EDGAR | EDGAR Company Search Results - Udemy, Inc. | |
| SV020 | Yahoo Finance | Udemy, Inc. (UDMY) Valuation Measures & Financial Statistics | |
| SV021 | CompaniesMarketCap | Udemy (UDMY) - Market capitalization | |
| SV022 | SEC EDGAR | EDGAR Company Search Results - Chegg, Inc. | |
| SV023 | Yahoo Finance | Chegg, Inc. (CHGG) Valuation Measures & Financial Statistics | |
| SV024 | CompaniesMarketCap | Chegg (CHGG) - Market capitalization | |
| SV025 | Oliver Wyman | How AI is reshaping SaaS valuations: a guide for investors | |
| SV026 | Federal Trade Commission | Generative AI Raises Competition Concerns | |
| SV027 | Grand View Research | AI Tutors Market Size, Share & Trends | Industry Report 2033 | |
| SV028 | MMR Statistics | Online Language Learning Market Insights 2025–2032 | |
| SV029 | Fortune Business Insights | Private Tutoring Market Size, Share & Industry Growth, 2034 | |
| SV030 | Sensor Tower | Speak: Language Learning - Apple App Store - US - Category Rankings, Keyword Rankings, Sales Rankings, Research, Performance, and Growth Metrics. |