录音转文字（Audio to Text）隐私政策
本政策说明「录音转文字」macOS App（以下简称「本 App」）处理哪些信息、为什么处理、保存多久，以及您可以怎么做。
1. 在这台 Mac 上转写
在 macOS 26 及以上选择「在这台 Mac 上」转写时：
- 录音、导入的音频和转写出的文稿只保存在您的 Mac 上，在本 App 的沙盒目录里。我们不接收、也看不到这些内容。
- 本 App 在这种模式下不联系我们的服务器。
- 语音识别由 macOS 自带的语音模型完成。第一次使用某种语言时，macOS 可能从 Apple 下载该语言的模型，这一过程由 Apple 处理。
2. 云端转写
只有在您选择云端转写之后，本 App 才会联系我们的服务器。第一次使用某个云端服务，以及它的数据条款有变化时，本 App 会先告诉您音频的接收方、处理地区和保留方式，您同意后才会上传。
我们会处理的信息：
- 设备标识：本 App 在您的 Mac 上生成一对密钥，私钥只保存在钥匙串里；我们的服务器保存对应的公钥，以及一个随机生成的设备编号和 App 版本号，用来确认请求确实来自这台设备。
- Apple 账户关联：本 App 会把 App Store 签发的「App 交易」凭据发给我们。它包含一个代表「哪个 Apple 账户取得了本 App」的编号，不包含您的姓名、邮箱或 Apple ID。我们用它发放一次性的试用时长，并让您购买的时长在同一 Apple 账户的其他 Mac 上也能使用。
- 购买记录：您购买时长时，付款由 Apple 处理，我们收不到任何付款信息。本 App 会把 Apple 签发的交易凭据发给我们，我们记录交易编号、商品和对应的时长，用于入账、防止重复入账和处理退款。
- App Store 地区：用来决定您所在地区可以使用哪些云端服务。
- 音频：您选择转写的录音会被切成若干段、压缩后，通过加密连接上传到我们的服务器，再转交给您选择的服务商识别。
- 转写结果：服务商返回的文字会在我们的服务器上短暂保存，直到本 App 取回。
- 用量记录：每次云端转写的时长、所用服务、语言、扣除的时长和时间。不包含音频或文字。
保存多久：
- 每一段音频在转写完成后立即从我们的服务器上删除。转写失败的分段会保留到您重试或删除，最长约 24 小时。
- 转写结果在本 App 取回、这次转写结束（完成、停止或删除）时删除；其余情况下约 24 小时后自动删除。
- 设备标识、Apple 账户关联、购买记录和用量记录会一直保存，用于计算您的剩余时长、处理退款和防止滥用。如需删除，请联系我们；删除后，剩余的时长也会一并作废。
服务商（音频的接收方）：
| 服务 |
接收方 |
处理地区 |
服务商如何处理 |
| Gemini |
Google |
Google 在各国的数据中心 |
不用这些内容改进产品，只为防止滥用而短期保留记录。上传到 Google 的文件在转写后立即删除 |
| Whisper |
Cloudflare |
Cloudflare 在各国的数据中心 |
识别在 Cloudflare 上完成，音频不交给其他公司；不用这些内容训练模型或改进服务 |
| 腾讯云语音识别 |
腾讯云 |
中国大陆 |
保存识别结果 24 小时，不用这些音频优化模型 |
您所在地区可用的服务以本 App 里显示的为准。各服务商的完整条款可在本 App 的同意窗口里打开。
我们的服务器与网络：我们的服务运行在 Cloudflare 上（Workers、D1、R2 与队列）。为了提供服务和防止滥用，Cloudflare 会处理 IP 地址等网络请求信息。我们的服务日志只记录请求的基本情况和错误，不记录音频、文稿内容、录音标题或文件路径。
跨境传输：依据您选择的服务，音频和文字可能在您所在国家或地区以外处理，处理地区见上表。
3. 我们不做的事
- 不要求注册账号，不收集姓名、邮箱、电话或通讯录。
- 不使用任何分析或广告工具，不做跨 App 或网站的追踪。
- 不出售任何信息，也不把信息用于广告。
- 不用您的音频或文稿训练任何模型。
4. 您的选择
- 选择「在这台 Mac 上」转写，音频就不会离开您的 Mac（需要 macOS 26 及以上）。
- 在本 App 里删除一条录音，会同时请求删除它在我们服务器上尚未清除的音频和文字；即使这一步没有完成（例如当时没有网络），服务器也会在约 24 小时后自动删除。
- 如需删除服务器上与您相关的记录，或对本政策有疑问，请通过支持渠道联系我们。
5. 录音须知
录音前请确认您有权录下相关的声音，必要时取得在场各方的同意。是否可以录音取决于您所在地的法律，与使用哪种转写方式无关。
6. 儿童隐私
本 App 不面向儿童，我们不会有意收集儿童的信息。
7. 政策更新
本政策变更时，我们会更新本页并修改生效日期。影响云端服务商或其数据处理方式的变更，本 App 会在下次上传前重新征求您的同意。
Audio to Text Privacy Policy
This policy explains what information the Audio to Text app for macOS (the "app") handles, why, for how long, and what you can do about it.
1. Transcribing on this Mac
When you choose to transcribe on this Mac, on macOS 26 or later:
- Recordings, imported audio and transcripts stay on your Mac, in the app's sandbox folder. We do not receive them and cannot see them.
- The app does not contact our servers in this mode.
- Recognition is done by the speech model built into macOS. The first time you use a language, macOS may download that language's model from Apple, which Apple handles.
2. Cloud transcription
The app contacts our servers only once you choose cloud transcription. Before audio is first sent to a cloud service, and again whenever that service's data terms change, the app shows you who receives the audio, where it is processed and how long it is kept, and uploads nothing until you agree.
What we handle:
- Device identity: The app creates a key pair on your Mac and keeps the private key in the Keychain. Our servers keep the matching public key, a randomly generated device ID and the app's version, to confirm that requests come from this device.
- Apple Account link: The app sends us the "app transaction" the App Store signs. It carries an identifier for the Apple Account that obtained the app. It does not contain your name, email address or Apple ID. We use it to grant the one-time trial, and to make the time you buy available on your other Macs signed in to the same Apple Account.
- Purchases: Apple processes payment, and we never receive payment details. The app sends us the transactions Apple signs. We record the transaction ID, the product and the time it grants, to credit it, to avoid crediting it twice, and to handle refunds.
- App Store region: Used to decide which cloud services are available where you are.
- Audio: A recording you choose to transcribe is split into parts, compressed, uploaded to our servers over an encrypted connection, and passed to the service you chose.
- Transcripts: The text a service returns is kept on our servers briefly, until the app retrieves it.
- Usage records: For each cloud transcription: its length, the service used, the language, the time charged, and when it happened. These records contain no audio or text.
How long it is kept:
- Each audio part is deleted from our servers as soon as it has been transcribed. A part that failed is kept until you try again or delete the recording, and for about 24 hours at most.
- Transcript text is deleted once the app has retrieved it and the transcription has ended (finished, stopped or deleted). Anything left otherwise is deleted automatically after about 24 hours.
- The device identity, Apple Account link, purchase records and usage records are kept to track your remaining time, handle refunds and prevent abuse. To have them deleted, contact us; any time left is then forfeited.
Services (who receives the audio):
| Service |
Recipient |
Where processed |
What the service does with it |
| Gemini |
Google |
Google data centers worldwide |
Does not use content to improve products and keeps logs briefly only to prevent abuse. Files uploaded to Google are deleted right after transcription. |
| Whisper |
Cloudflare |
Cloudflare data centers worldwide |
Recognition runs on Cloudflare; audio goes to no other company. Cloudflare does not train models or improve services with it. |
| Tencent Cloud |
Tencent Cloud |
Mainland China |
Keeps recognition results for 24 hours; does not use audio to improve models. |
The services available in your region are the ones the app shows. Each service's full terms can be opened from the app's consent window.
Our servers and the network: Our service runs on Cloudflare (Workers, D1, R2 and Queues). To provide the service and prevent abuse, Cloudflare processes network information such as IP addresses. Our service logs record the basics of each request and any errors. They never record audio, transcript text, recording titles or file paths.
International transfers: Depending on the service you choose, audio and text may be processed outside your country or region. The table above shows where.
3. What we don't do
- We don't ask you to create an account, and we don't collect names, email addresses, phone numbers or contacts.
- We use no analytics or advertising tools, and we don't track you across apps or websites.
- We don't sell any information or use it for advertising.
- We don't train any model on your audio or transcripts.
4. Your choices
- Transcribe on this Mac, and your audio never leaves your Mac (requires macOS 26 or later).
- Deleting a recording in the app also asks our servers to delete whatever of its audio and text they still hold. Even if that request does not get through, for instance without a network connection, the servers delete it automatically after about 24 hours.
- To have records about you deleted from our servers, or with questions about this policy, contact our support team.
5. About recording
Before you record, make sure you are allowed to record the people and sounds involved, and ask for consent where needed. Whether you may record depends on the law where you are, not on how the recording is transcribed.
6. Children
The app is not directed to children, and we do not knowingly collect information from them.
7. Changes
When this policy changes, we update this page and its effective date. If a change affects a cloud service or how it handles data, the app asks for your consent again before the next upload.