隐私政策

Privacy Policy

港粤语 Canspeak · 生效日期：2026 年 9 月 23 日

English version is available below.

本隐私政策说明「港粤语 Canspeak」（下称「本 App」）如何收集、使用和保护你的信息。 使用本 App 即表示你同意本政策所述的做法。

一、我们收集的信息

为提供粤语学习功能，本 App 可能收集以下信息：

  • 麦克风音频：当你使用「发音评分」或「语音输入」时，我们会录制你的语音， 用于发音打分和语音转文字。音频会传输到后端评测服务处理；除你另行授权用于模型训练的情形 （见第二节）外，处理完不作长期保存，也不用于身份识别。
  • 账号信息：如你选择登录，我们会收集邮箱或手机号，用于创建和识别你的账号、 保存学习进度。设备会生成一个匿名标识用于绑定游客数据。
  • 学习数据：你的学习进度、打卡记录、对话记录、发音评分结果，用于同步和展示你的进度。
  • 你主动输入的内容：翻译文本、与 AI 角色的对话内容，用于生成回复。

二、模型训练数据的收集（一次性授权）

为改进发音评测、语音识别、对话与翻译模型，我们提供「Canspeak 改进计划」， 邀请你授权我们将相关数据用作训练样本。对此我们遵循以下规则：

  • 一次性授权，覆盖四个场景：你首次使用发音评分、语音输入、AI 对话或翻译中任意功能后，App 会展示一次授权页， 逐项列明四个场景各自收集的数据内容与用途。你点击「同意并加入改进计划」后， 即完成对全部四个场景的授权；之后使用这些功能时，数据将自动用于模型训练，不会再次弹窗确认。 拒绝不影响你正常使用任何功能。
  • 授权范围与用途：
    • 发音评分——收集你的录音与发音评分结果，用于训练发音评测模型；
    • 语音输入——收集你的录音与转写文本，用于训练粤语语音识别模型；
    • AI 对话——收集你与 AI 角色的对话内容，用于训练粤语对话模型；
    • 翻译——收集你输入的文本与翻译结果，用于训练翻译模型。
  • 服务器端统一收集：经你授权的数据会加密传输到我们的服务器统一保存，客户端不直接写入训练数据库。
  • 去标识化：训练样本使用随机编号保存，不包含你的邮箱、手机号等身份信息； 设备标识经不可逆哈希处理，仅用于数据去重。
  • 七天自动删除：所有训练样本自收集起最长保存 7 天，到期由系统自动物理删除（包括音频文件），并保留删除审计记录。
  • 不会分享给第三方用于训练：经授权收集的训练数据仅在我们受控的环境中使用，不会提供给微软 Azure、Google Gemini 或任何其他第三方用于其模型训练。
  • 随时退出：你可以在「我的 → 隐私」页随时关闭「参与改进计划」，关闭后立即停止收集，无需任何其他操作。 撤回不影响撤回前已收集数据的处理；已收集数据仍会在 7 天期限内自动删除。

需要说明的是：由于训练样本已去标识化，我们无法在样本库中定位并删除属于某一特定用户的单条样本； 7 天自动删除机制确保你的数据不会长期留存。删除账号时，你的授权记录与未过期样本会一并清除。

三、信息如何使用

  • 提供并改进翻译、对话、发音评分、语音识别等核心功能
  • 在你一次性授权的前提下，将四个场景的数据用于模型训练（见第二节）
  • 保存并同步你的学习进度与账号资料
  • 保障服务安全、防止滥用（如接口限流）

我们不会将你的个人信息出售给第三方，不会用于跨 App／网站的广告追踪。

四、第三方服务

为实现相关功能，本 App 会将必要数据发送给以下服务商处理：

  • 语音评测 / 语音合成（微软 Azure 语音服务）：处理发音评分与朗读，需发送音频或文本。
  • AI 对话 / 翻译（Google Gemini 大模型服务）：处理你的对话与翻译请求。
  • 后端与账号服务（Supabase）：存储账号与学习数据、处理请求；经你授权的训练数据也保存在该服务的独立加密存储中。
  • Apple：订阅购买通过 Apple 内购处理，我们不会接触你的支付卡信息。

上述服务商各自的隐私政策适用于其处理的数据。 发送给上述服务商的数据仅用于向你提供当次功能，我们不会将训练数据集提供给它们。

五、数据存储与安全

我们采用合理的技术与管理措施保护你的信息。数据通过加密网络（HTTPS）传输； 训练数据与账号数据分开存储，并施以更严格的访问控制。 尽管如此，任何网络传输或存储都无法保证绝对安全。

六、你的权利：删除账号与数据

你可以随时在 App 内「我的 / 账号」页发起删除账号， 删除后你的账号信息与学习数据将被清除且无法恢复， 你的训练数据授权记录与未过期样本也会一并删除。 如有疑问也可通过下方邮箱联系我们协助处理。

七、儿童隐私

本 App 面向 12 岁及以上用户。我们不会有意收集低龄儿童的个人信息， 也不会邀请未成年人参与模型训练数据授权。

八、政策更新

我们可能会不时更新本政策，更新后会修改本页顶部的生效日期。重大变更会在 App 内提示。 政策更新后，你此前作出的训练数据授权自动失效， 我们会在你下次使用相关功能时重新向你说明并请求授权。

九、联系我们

如对本隐私政策有任何疑问，请联系：
support@canspeak.uk

Privacy Policy

Canspeak · Effective Date: September 23, 2026

This Privacy Policy explains how "Canspeak" (the "App") collects, uses, and protects your information. By using the App, you agree to the practices described in this policy.

1. Information We Collect

To provide Cantonese learning features, the App may collect the following information:

  • Microphone audio: When you use pronunciation scoring or voice input, we record your voice to score pronunciation and convert speech to text. Audio is sent to backend assessment services for processing; unless you separately authorize it for model training (see Section 2), it is not stored long term after processing and is not used to identify you.
  • Account information: If you choose to sign in, we collect your email address or phone number to create and identify your account and save learning progress. A device may generate an anonymous identifier to link guest data.
  • Learning data: Your learning progress, check-in records, conversation history, and pronunciation scores are used to sync and display your progress.
  • Content you enter: Translation text and conversations with AI characters are used to generate responses.

2. Collection of Model Training Data (One-Time Authorization)

To improve our pronunciation assessment, speech recognition, dialogue, and translation models, we offer the "Canspeak Improvement Program" and invite you to authorize the use of related data as training samples. We follow these rules:

  • One-time authorization covering four scenarios: The first time you use any of pronunciation scoring, voice input, AI chat, or translation, the App presents a single authorization page that itemizes what data each of the four scenarios collects and why. Tapping "Agree and Join the Improvement Program" authorizes all four scenarios at once. From then on, data from these features is used for model training automatically, with no further pop-up confirmations. Declining never affects your access to any feature.
  • Scope and purposes of authorization:
    • Pronunciation scoring — your recordings and pronunciation scores, used to train the pronunciation assessment model;
    • Voice input — your recordings and transcribed text, used to train the Cantonese speech recognition model;
    • AI chat — your conversations with AI characters, used to train the Cantonese dialogue model;
    • Translation — the text you enter and its translation results, used to train the translation model.
  • Unified server-side collection: Authorized data is transmitted over encrypted connections and stored centrally on our servers. The app client never writes to the training database directly.
  • De-identification: Training samples are stored under random identifiers and do not contain your email, phone number, or other identity information. Device identifiers are irreversibly hashed and used only for deduplication.
  • Automatic deletion after 7 days: Every training sample is kept for a maximum of 7 days from collection, after which it is automatically and permanently deleted (including audio files). Deletion audit logs are retained.
  • Never shared with third parties for training: Training data collected with your authorization is used only in environments under our control. It is never provided to Microsoft Azure, Google Gemini, or any other third party for their model training.
  • Opt out anytime: You can turn off "Join the Improvement Program" at any time under "Me → Privacy". Collection stops immediately with no further action required. Withdrawal does not affect the processing of data collected before withdrawal; that data is still automatically deleted within the 7-day period.

Please note: because training samples are de-identified, we cannot locate or delete a specific individual sample within the sample library. The 7-day automatic deletion mechanism ensures your data is never retained long term. When you delete your account, your authorization records and any unexpired samples are deleted as well.

3. How We Use Information

  • Provide and improve core features such as translation, chat, pronunciation scoring, and speech recognition
  • With your one-time authorization, use data from the four scenarios for model training (see Section 2)
  • Save and sync your learning progress and account information
  • Protect service security and prevent abuse, such as API rate limiting

We do not sell your personal information to third parties, and we do not use it for advertising tracking across apps or websites.

4. Third-Party Services

To provide related features, the App may send necessary data to the following service providers:

  • Speech assessment / text-to-speech (Microsoft Azure Speech Services):Processes pronunciation scoring and read-aloud features, which may require sending audio or text.
  • AI chat / translation (Google Gemini large language model services):Processes your chat and translation requests.
  • Backend and account services (Supabase):Stores account and learning data and handles requests. Training data you authorize is also stored in a separate, access-controlled storage area within this service.
  • Apple: Subscriptions are processed through Apple In-App Purchase. We do not access your payment card information.

Each service provider's own privacy policy applies to the data it processes. Data sent to these providers is used only to deliver the requested feature; we do not provide our training dataset to them.

5. Data Storage and Security

We use reasonable technical and administrative measures to protect your information. Data is transmitted over encrypted connections (HTTPS). Training data is stored separately from account data under stricter access controls. Even so, no method of network transmission or storage can be guaranteed to be completely secure.

6. Your Rights: Account and Data Deletion

You can request account deletion at any time in the App under "Me / Account". After deletion, your account information and learning data will be removed and cannot be restored; your training-data authorization records and any unexpired samples will be deleted as well. If you have questions, you may contact us at the email address below for assistance.

7. Children's Privacy

The App is intended for users aged 12 and above. We do not knowingly collect personal information from younger children, and we do not invite minors to participate in model training data authorization.

8. Policy Updates

We may update this policy from time to time. When we do, we will update the effective date at the top of this page. Material changes may also be announced in the App. After a policy update, your previous training-data authorization lapses automatically, and we will explain the changes and request your authorization again the next time you use a related feature.

9. Contact Us

If you have any questions about this Privacy Policy, please contact:
support@canspeak.uk