Trace Voice

隐私政策

Trace Voice / Apollo Project / 最后更新:2026年11月19日

处理的信息

Trace Voice 提供语音输入、文字润色、录音保存、转写和会议纪要生成功能。无障碍功能会确认输入框和麦克风使用状态,并将结果填入输入框。密码输入框不在支持范围内。录音在用户操作且授予麦克风权限后开始。

为管理账户和了解使用情况,我们会为每个已登录账户记录所用设备类型(Windows、Mac、iPhone、Android)、应用版本、国家/地区、电子邮件地址以及首次和最近一次使用时间。删除账户时这些记录会被清除。

在应用中录制的音频会经由 Trace Voice 服务器发送到语音识别服务进行转写。在 Android 上,如果关闭了音频保存且只使用一种语言,也可能使用设备上设置的语音识别服务(通常为 Google),并适用该服务的条款。整理现有语音输入或所选文字时,会发送相应文字和词典。

录音保存、转写与会议纪要

当前版本的应用会将您通过 Trace Voice 输入的文字自动保存在设备内。开启“自动保存语音”后,语音也会一并保存。已保存的语音、转写和会议纪要会在设备内加密存储。语音默认在 30 天后删除,可选择 7 天、30 天、90 天或不设期限。设置更改将适用于之后的新语音。关闭语音保存后,已有录音不会自动批量删除。如需删除不需要的项目,请在录音页面中删除。录音过程中的临时文件存放在应用专用区域,确认保存后再加密。

重新识别已保存语音或转写导入的语音时,语音会经由 Trace Voice 服务器发送至语音识别服务。生成带时间戳的字幕时,语音会发送至语音识别服务。生成会议纪要时,转写文字会经由服务器发送至 AI 服务。为避免重试时重复处理,处理结果会加密暂存,同一账户可在 24 小时内重新获取。过期后将定期删除(通常在 48 小时内)或在删除账户时清除。Trace Voice 的数据库不提供将处理用录音长期保存的功能。导出的文件由用户自行选择的保存或分享位置管理,不会与应用内的删除操作联动。

发送目的地与用途

待处理的文字和个人词典会经由 Trace Voice 服务器发送至外部 AI 服务。发生故障时,可能会发送至其他 AI 服务。登录使用 Google Firebase Authentication,处理电子邮件地址、用户标识符和认证信息。选择 Google 登录时,Google 账户的显示名称等基本资料也会由 Firebase 处理,并用于账户管理。密码由 Firebase 管理,不会发送至 Trace Voice 的 API 服务器。各服务的处理与保存条件同样适用。

通信采用 HTTPS 加密。我们不进行广告投放,也不为广告目的进行追踪。

Google Play 购买信息

进行购买时,为了确认和恢复购买以及防止不当重复使用,我们会在服务器保存 Google Play 购买令牌、账户标识符的哈希值、套餐、有效期和确认时间。信用卡号等支付信息由 Google Play 处理,Trace Voice 不会获取。即使删除账户,购买记录仍会为防止不当重复使用而保留。如需咨询购买信息的删除,请通过下方联系窗口提出。仅删除账户并不会取消 Google Play 的定期订阅,请另行在 Google Play 中取消。

App Store 购买信息

在 iPhone 或 iPad 版中通过 App Store 进行定期订阅时,为了确认和恢复购买以及防止不当重复使用,我们会在服务器保存 Apple 的原始交易 ID、账户标识符的哈希值、商品和套餐、购买日期、有效期、状态和确认时间。为将购买与账户关联,我们会向 Apple 提供由账户生成的识别值(不包含电子邮件地址等信息)。收到 Apple 发送的购买状态通知时,我们会先向 Apple 确认,再更新记录。信用卡号等支付信息由 Apple 处理,Trace Voice 不会获取。删除账户时,我们会删除交易 ID 和购买详情,为防止不当重复使用,会保留交易 ID 的哈希值、账户标识符的哈希值、购买环境(正式或测试)以及删除时间。仅删除账户并不会取消 App Store 的定期订阅,请在 iPhone 的“设置”(Apple 账户 → 订阅)中取消。

在网站和 Windows 版中的购买(Stripe)

在网站或 Windows 版购买方案时,付款由 Stripe 处理。为了管理购买和订阅,我们会把 Trace Voice 账户标识和电子邮件地址提供给 Stripe,并在服务器上保存 Stripe 客户 ID、方案、订阅状态和当前周期的结束日期。卡号等付款信息由 Stripe 处理,Trace Voice 不会获取。Stripe 自身的处理和保存条款同样适用。删除账户时,我们会请求 Stripe 删除客户信息,订阅随即终止,剩余期间将无法使用。服务器上的购买记录也会删除。

Homophone data: JMdict, © Electronic Dictionary Research and Development Group, CC BY-SA 4.0.

Kanji reading data: KANJIDIC2, © Electronic Dictionary Research and Development Group, CC BY-SA 4.0.

说话方式学习

When enabled in Settings, the app learns frequently used words and words you correct, on your device, from the text of memos stored there. Audio is not used to learn these words. It also remembers, on your device and in the same way, word pairs where the server corrected a recognition mistake, once the same pair has appeared in two dictations (words only). Learned words are encrypted on your device and excluded from backups. They are sent with recognition and text-finishing requests, like dictionary words, and are not stored on our servers. Words containing numbers or email addresses are not learned. You can review, erase or disable learning in Settings, and deleting a memo discards what was learned from it.

发送录音

Only when you agree to contribute recordings (on Android, together with agreeing to personalize speech recognition; on Windows, as a separate agreement), we receive the audio and text of the recordings you choose, to use in considering improvements to speech recognition. Automatic sending of new recordings, about an hour later, happens on Android together with that agreement, and on Windows only if you give a separate agreement for it. On Windows, only recordings made with this PC's microphone are sent; imported audio and the dictation audio kept for voice analysis are not sent. You can exclude any recording and stop at any time. Recordings found to contain little of your own voice are not sent. What is sent is encrypted on your device and cannot be decrypted on the server. The decryption key is kept offline, separately from the server, and only Trace Voice can decrypt it. It is stored encrypted on our servers and deleted automatically after 30 days. When you stop, recordings already sent are deleted from the server. They are also deleted if you delete your account. It is not provided to third parties.

地区设置

开启语音个性化并设置国家、地区或方言后,这些信息会在每次语音输入时与音频一起用于处理,并传给 AI 服务,作为识别口音、方言和地名的线索。不使用 GPS。该设置保存在设备上;开启同步时,也会加密保存在 Trace Voice 服务器上,供同一账户的其他设备使用。关闭后即被删除。

推荐计划

登记推荐码时,我们会保存推荐人与使用者标识符的哈希值、推荐日期时间、条件达成情况,以及优惠字数、使用量和有效期,用于发放优惠和防止不当重复领取。推荐人可查看人数等汇总结果,被推荐人的电子邮件地址和输入内容不会公开。对于推荐合作伙伴,我们还会保存对应订单的标识符、销售核对结果和佣金记录。推荐及优惠、佣金记录将为结算和防止不当行为而在删除账户后继续保留。如需咨询删除或保存期限,请通过下方联系窗口提出。

关于 Windows 版

Windows 版会将录音和词典发送至服务器,进行语音识别和文字润色。文字历史记录和词典保存在电脑上,保存期限可在设置中更改。未能处理的录音会保留在电脑上以便重试,7天后自动删除。

On Windows, recordings are stored encrypted on this PC, and by default only the audio is deleted after 30 days. To transcribe, audio is sent through the server to a speech recognition service; to create captions, audio is sent to a speech recognition service. For meeting notes and polishing selected text, text is sent through the server to AI providers. When enabled, your speaking region and learned words are included in requests in the same way as the dictionary. If you give an additional agreement, the audio of your own short dictations is also kept encrypted on this PC for up to 30 days for voice analysis (see “Voice analysis” below), and if you agree to contribute recordings, recordings made with this PC's microphone are sent encrypted (see “Sending recordings” above).

关于 Mac 版和 iPhone・iPad 版

Mac 版使用麦克风和辅助功能权限,后者用于把完成的文字输入到当前的输入位置。iPhone・iPad 版键盘的“完全访问”用于把 Trace Voice 应用中整理好的文字传给键盘。两者的音频都会经由 Trace Voice 服务器发送到语音识别服务,文字的整理方式与其他设备相同。

声音分析

开启语音个性化并开始声音分析后,系统会请您朗读所显示的句子,并将该音频经由服务器发送至语音识别服务和外部 AI 服务,用于分析口音、发音倾向和方言用词。音频不会保存在 Trace Voice 的服务器上。分析结果会加密保存,会以您说话方式的摘要形式向您显示,同时作为识别和文字润色的参考线索。为识别您本人的声音,会将数秒的音频加密后仅保存在该设备上。此音频不会同步。

If voice analysis has been performed, parts of recordings of 20 seconds or longer are sent through the same route, up to a few per day, to check how much of the audio is your own voice. Recordings with little of your own voice are excluded from learning and from recording submission. The analysis results are not periodically rewritten from recordings; the saved analysis stays as it was created from what you read aloud.

Only if you give an additional agreement (on Android it is included in agreeing to contribute recordings; on Windows it is a separate agreement), parts of your own short dictations are also sent through the same route, up to a few per day, to check whether they are your voice, and up to one dictation per day of 3 to 40 seconds that is confirmed to be your voice is sent through the same route to analyze differences in pronunciation. Only sound combinations confirmed repeatedly across several recordings and several days are added, as hint text, to recognition and text-finishing requests, and checks older than 60 days are not used. Audio sent for this analysis is also not stored on Trace Voice's servers. On Windows, the audio of the relevant dictations is kept encrypted on this PC for this purpose. It is kept for at most 30 days; if your audio retention setting is shorter, that period applies, and even if you chose “No limit” it is deleted after 30 days. This audio is not shown in history or the recordings list, and is neither synced nor sent as a recording. If you turn personalization off or erase what was learned, the analysis results, the stored voice, the information collected for this analysis and the kept audio, and the voice-check results are deleted.

同步

登录后,词典、语言、文体等设置,以及您的说话地区和声音分析结果,将经加密保存在 Trace Voice 的服务器上,以便在同一账户的设备上使用。仅当您开启音频同步时,录音音频才会经加密保存在 Trace Voice 的服务器上,并在 30 天后删除。用于同步的音频不会发送给 AI 服务。加密密钥由服务器端管理,并非端到端加密。您可以在设置中关闭此同步。所说文字的同步默认关闭,仅在您开启时才会加密保存。已同步的数据可在设置中删除,删除账户时也会一并清除。

保存与删除

History is stored encrypted on the device, keeping the latest 30 items, and the dictionary is also stored on the device. Except for the temporary retry storage and sync described above, Trace Voice's database does not store the text or dictionary being processed. Aggregates of regular monthly usage counts and input character counts are retained for up to 31 days. On Android and Windows, for the cumulative dictation time shown in History, we store, per account, a per-device identifier and two numbers (the time spent dictating and the estimated time it would take to type it by hand; no text or audio) and keep them until the account is deleted. Referral reward balances and usage records, and cumulative character counts used for eligibility checks, are retained separately according to the referral program above. To help prevent fraudulent registrations, restriction records including the IP address at registration are temporarily retained.

作为上述保存政策的例外,在语音输入、重新识别已保存语音和生成会议纪要时,为防止连接中断后重复处理,会将完成结果加密暂存。仅限同一账户的同一处理,可在 24 小时内重新获取;过期后将定期删除(通常在 48 小时内)或在删除账户时清除。录音本身和词典不会被保存。

可以通过应用内的“删除账户”,或账户管理页面进行删除。为阻止已删除账户的访问,我们会在删除完成后最长7天保留标识符的哈希值。若通过网页删除,请也在设备端删除设备内的词典和历史记录。未完成的删除可以重试。

联系我们

Trace Voice 运营方:info@traceforai.com