| 类型 | 具体内容 | 用途 |
|---|---|---|
| 语音 | 你按下麦克风后说的话（音频） | 转成文字，判断你说了什么 |
| 文本 | 识别出的你的发言、AI 的回复 | 维持对话上下文、生成练习记录 |
| 练习记录 | 练习时长、完成剧本、打卡天数 | 展示你的学习进度 |
| 本地标识 | 一个随机字符串（存在你浏览器本地） | 识别「这是同一个人」，以便保存进度 |
我们不收集：通讯录、位置、相册、剪贴板、其他 App 数据， 以及其他任何与英语练习无关的信息。
不录音留存：语音在你说话停顿后即被送往识别服务转为文字， 我们不保存你的原始录音；服务端只保留识别出的文字与练习统计。
本内测版本的服务器部署在 中国香港（阿里云）。 这意味着你的练习数据会被传输并存储在中国香港。
如你对数据出境有顾虑，可以随时停止使用并联系我们删除你的全部数据。
为了提供语音识别与 AI 回复，我们需要把相应内容发送给以下服务商处理：
| 服务商 | 处理内容 | 用途 |
|---|---|---|
| Deepgram | 语音音频 | 语音转文字（ASR） |
| 微软语音服务 | AI 回复文本 | 文字转语音（TTS） |
| Groq | 文本对话内容 | 生成 AI 的英语回复（主用） |
| Google Gemini | 文本对话内容 | 主用模型失败时，接着生成 AI 的英语回复 |
| DeepSeek | 文本对话内容 | 前两家都失败时，接着生成 AI 的英语回复 |
上述服务商仅在其服务范围内处理数据，我们不会将你的数据出售或用于广告。
如果使用者是不满 14 周岁的儿童，按《个人信息保护法》与 《儿童个人信息网络保护规定》，我们需要父母或其他监护人同意后才开始处理。
会处理的信息：
不收集：真实姓名、学校名称、住址、位置、通讯录、相册， 以及任何与服务无关的信息；AI 也不会追问这些。
存储与删除：数据存放在上述香港服务器，保留至内测结束。 监护人可随时联系我们查看、更正或删除孩子的全部数据， 也可以要求撤回同意 —— 撤回后我们不再处理，并删除相应数据。
撤回同意（应用内可直接操作）：在应用里点阶段旁的 「设置」，弹窗底部会出现「撤回监护人同意」按钮， 点两下即可（第二下是防误触的确认）。撤回后：孩子会 立即回落为面向成年人的默认阶段，我们不再按少儿档处理其信息； 孩子原先选择的阶段会被保留，重新勾选同意即可恢复，不必重新选择。 不愿在应用内操作时，也可通过下方「联系我们」提出。
投诉与举报：可通过下方「联系我们」提出，我们会尽快处理。
不同意以上内容时，请选择「暂不使用这个阶段」；这不影响使用面向成年人的其他阶段。
对话中的 AI 语伴由大语言模型生成，不是真人。 它的表达可能不准确或不符合真实语境，请勿将其作为权威的语言或事实依据。 界面中始终显示「由 AI 生成」标识，且无法关闭。
内测期间，你的练习记录会一直保留以便你查看进度。 内测结束或你提出要求时，我们会删除对应数据。
数据相关请求（查看 / 删除 / 疑问），请联系内测发起人：
aipoden.com 对应的内测联络人。
本政策如有变更，我们会在应用内提示。继续使用即表示接受更新后的版本。
Audio of your speech (only while you hold the mic), the transcribed text of your turns and the AI's replies, your practice statistics (duration, scenarios, streak), and a random local identifier stored in your browser. We do not collect contacts, location, photos, clipboard, or data from other apps. Raw audio is not retained — it is sent to the speech service for transcription and only the resulting text is kept.
The beta server is hosted in Hong Kong, China (Alibaba Cloud). Your practice data is transferred to and stored there.
To provide speech recognition and AI replies, we send the relevant content to these processors:
| Provider | What they process | Purpose |
|---|---|---|
| Deepgram | Speech audio | Speech-to-text (ASR) |
| Microsoft speech services | AI reply text | Text-to-speech (TTS) |
| Groq | Conversation text | Generate the AI's English replies (primary) |
| Google Gemini | Conversation text | Generate replies if the primary model fails |
| DeepSeek | Conversation text | Generate replies if the first two both fail |
These providers process data only within their service scope. We never sell your data or use it for advertising.
If the user is a child under 14, we require consent from a parent or guardian before processing their data, as required by China's Personal Information Protection Law and the Provisions on Cyber Protection of Children's Personal Information.
What we process: nickname and dialogue text (to run the conversation and record progress); voice, sent in real time to a speech-recognition service — we do not store audio files; an optional contact detail used only for beta follow-up.
What we never collect: real names, school names, home address, location, contacts, photos, or anything unrelated to the service. The AI will not ask for them either.
Storage and deletion: data is stored on the Hong Kong server named above until the beta ends. A guardian may contact us at any time to access, correct or delete the child's data, or to withdraw consent — after which we stop processing and delete the corresponding data.
Withdrawing consent (in-app): tap “Settings” next to the stage selector; a “Withdraw guardian consent” button appears at the bottom of the panel. Two taps complete it (the second is a confirmation, to prevent accidental withdrawal). After withdrawal the child immediately drops back to the adult default stage and we stop processing their information as a child; the stage they chose is kept, and ticking the consent box again restores it without re-selecting. You may also contact us using the details below.
If you do not agree, simply choose "not now" for that learning stage; the adult stages remain available.
Your conversation partner is generated by a large language model, not a human. Its output may be inaccurate. The "AI-generated" label is always shown in the interface and cannot be turned off.
Practice records are kept during the beta so you can track progress. You may request access to or deletion of your data at any time by contacting the beta organizer. Deletion is irreversible.
← 返回练习