Respeak Privacy Policy

Version: 2026-11-06.2 · Effective: November 6, 2026 · Operator: xiang zhang

This English version is a translation of the Chinese Privacy Policy provided for convenience. If the two versions differ, the Chinese version prevails.

Respeak (www.getspeakeys.com, the "Service", "we") is operated by xiang zhang. The Service helps you fix individual sentences in your recordings: it transcribes them, generates or imports replacement speech, and splices the replacement back into the recording; it can also remove fillers, overlong pauses and unwanted words by editing the transcript (Remove fillers). We process your personal information in accordance with the Personal Information Protection Law of the People's Republic of China and other laws and regulations. Please read this policy carefully before using the Service, especially the sections in bold on sensitive personal information.

1. What information we collect

1. Account information

• Email sign-up: email address, display name, password (stored only as a salted scrypt hash; we cannot know the plain text) and whether the email has been verified.
• Email verification and password reset: we email you a one-time link; we store only an irreversible hash (SHA-256) of the random token in the link, its purpose and its validity (verification links 24 hours, password-reset links 30 minutes, void once used). To prevent abuse, we count requests per email address and IP address (hashed) for about 15 minutes.
• Third-party sign-in (Google, and Microsoft when enabled on the server): the user ID, display name and email returned by the platform. We do not store these platforms' access tokens. When you sign in with Google in the iOS app, the sign-in happens in a system web window; we then give the app a one-time sign-in code (we store only its hash; valid for 2 minutes, single use), which the app exchanges for a session.
• Sign in with Apple (website and iOS app): the user ID, email address and whether it is verified (if you choose “Hide My Email”, a relay address provided by Apple; system emails we send there are forwarded to you by Apple), and the name you chose to share on first authorization (used as the display name). So that we can revoke your Apple authorization for this service when you delete your account, we also store the refresh token Apple issues, encrypted, and use it for nothing else.
• Legacy shared workspace (access passphrase): no personal identity information is collected; only the session is recorded.
• Agreement records: the version of this policy and the Terms of Use you agreed to, and when.

2. Recordings and edits

• The recordings you upload (converted on the server to a 48 kHz mono WAV working file), the audio versions created while editing (the last 20 steps can be undone), the transcripts and your corrections, and, when you use Remove fillers, the recognized timing of every word.
• The new sentences you type, the replacement clips generated or imported, and the exported WAV/MP3 files.
• Recordings may contain your voice or other people's voices and conversations, which may include other personal information; please upload only recordings you are entitled to process.

3. Voice (voiceprint) — sensitive personal information

A voice can identify a specific natural person and is sensitive personal information. When you use "original voice", we send the audio of the sentence you selected and a few neighbouring sentences (up to 30 seconds in total), to the speech-synthesis provider Alibaba Cloud Model Studio (阿里云百炼) as a reference to generate speech that sounds like that voice. To shorten the wait, once you have given voice consent we send the reference as soon as you start changing a sentence, rather than when you press Generate. Alibaba Cloud Model Studio registers a temporary voice from the reference so later sentences of the same recording can use it; the voice is kept with the recording, and we delete it when the recording is deleted (3 days after its last edit, 7 days for accounts that have bought or redeemed credits), when the voice consent is withdrawn, or when it expires (checked every 10 minutes). Original voice may only be used with your own voice. Before it is used for the first time on each recording, we obtain your separate consent in a dedicated "Voice consent" window, where you confirm that the reference voice is your own; other people's voices cannot be used. With a personal account, you may additionally tick to have this confirmation cover all recordings you upload to this account for 30 days (otherwise it covers only the current recording). You can withdraw consent at any time in "Voice consent". We do not build a separate voiceprint database and do not use your voice to train models.

4. Voice-consent records

To prove that separate consent was obtained, we keep: the consent ID, account, recording project, consent type (own voice) and scope (current recording / all recordings in the account), withdrawal time, the version of the consent text, the confirmation time, the IP address and browser information (User-Agent).

5. Compliance audit logs

To meet the requirements for labelling AI-generated synthetic content, we record: AI speech-generation events (account, project ID, generation ID, provider, model, whether the original voice was cloned, consent ID), exports containing AI content (content ID ProduceID, whether the audio cue was added), your confirmation when exporting the "version without the audio cue" from the Chinese interface (account, time, project, audio version, export ID, IP), agreement records, account-deletion events (account ID, time and number of projects deleted), password resets by email (account ID, time, IP) and email verifications (account ID, time), and, once paid features are enabled, credit redemptions (account ID, code batch and number of credits, not the code itself), online purchases credited and refunds deducted (account ID, order number, credits), and credits granted or refunded by an administrator. Audit logs contain no transcripts, new sentences or audio of any kind.

6. Operating logs and local storage

• Service logs: request ID, API path, status code, duration and similar, for troubleshooting and security; they do not record transcripts, audio or keys.
• Sign-in protection: to limit brute-force attempts we briefly (about 15 minutes) keep hashed counts per IP address (IPv6 by /64 prefix) and sign-in email; for Sign in with Apple, a hash of the one-time random value (nonce) in the sign-in request is kept for about 10 minutes so that the same sign-in credential cannot be used twice.
• Cookies: only a necessary HttpOnly session cookie (valid for 7 days) and a one-time state cookie during third-party sign-in; no advertising or analytics cookies.
• Visitor statistics: we use Umami, an open-source tool we host ourselves, to count page views, referring websites, browser types and approximate region; it uses no cookies, does not store IP addresses, the data is not shared with third parties, and it is kept for no more than 2 years.
• Browser local storage: only the ID of the recording you opened most recently (and your interface-language choice).

7. Credits and redemption records (once paid features are enabled)

• Credit balance and credit lots: the source (free, purchase, redeem code, granted), amount, remaining credits and expiry date of each lot.
• Credit history: the amount of each change, the reason (free credits, purchase, redemption, replacement, regeneration, returned, granted, refund, deduction, code revoked, refund deducted), the related recording project ID and generation ID, the balance after the change and the time.
• Redemption records: the batch of each code you redeemed and when (and, if a code was disabled or revoked, when). We store only an irreversible hash (HMAC-SHA256) of redeem codes, never the codes themselves.
• Usage metering: per recording, the number of generations with the server speech service since the last confirmed replacement (and how many of them were beyond the free allowance); per day, the free generations used and the transcription time; to prevent code guessing, redemption attempts per account and IP address (hashed) for about 1 hour; checksums of audio generated with the server speech service (kept 30 days, without the audio itself), to detect generated speech being imported as your own recording to avoid charges and AI labels.
• Purchase orders: when you buy a credit pack online we record a randomly generated order number, the account, the pack, the number of credits, the order status, and from Creem the checkout ID, order ID, customer ID, amount and currency paid, and the amounts and times of payments and refunds (chargebacks); to prevent abuse, we count checkout pages created per account for about 1 hour. Payment information is collected and processed by Creem; we do not receive your card, payment-account or other payment details; we do not store the buyer's email, name or country that Creem includes in its payment notifications.
• In-app purchases: when you buy a credit pack in the iOS app, we receive the transaction signed by Apple and keep its transaction ID (and original transaction ID), product, quantity, amount and currency, country or region (App Store storefront), purchase time, environment (production or test), and the times Apple refunds the purchase or reverses a refund. To match purchases to your account, we create a random identifier for the account, which the app passes to Apple when you buy and which is deleted when you delete your account. To prevent abuse, we count purchase submissions per account for about 1 hour. We do not receive or store your Apple ID, name, email or payment details.

2. How we use this information

• To provide the core features: sign-in, storing recordings, transcription, generation, splicing, undo and export;
• To meter and charge credits, handle redemptions and refunds, prevent abuse, and keep financial records (once paid features are enabled);
• To meet legal obligations such as labelling AI-generated synthetic content, obtaining separate consent for deep synthesis and keeping records;
• To keep accounts and the Service secure, and to handle complaints, disputes and regulatory requests.

We do not use your information for targeted advertising, do not sell your personal information, and do not use recordings to train any model.

3. Processors and third-party services

Some features require sending data to the following service providers; we send only the minimum needed for the feature:

Before your first upload (or before a recording uploaded earlier is first transcribed, trimmed with Remove fillers or used to generate), we ask for your separate consent to send recordings to Alibaba Cloud Model Studio; without it nothing is sent, and transcription and new sentences are not available.

The speech service is always called with our server key, at our cost. Each provider handles the data it receives under its own privacy policy and terms. Third-party sign-in and online purchases redirect you to platforms outside China, which handle the related information under their own policies.

4. Where data is stored and for how long

• Data is stored on the Tencent Cloud servers we rent, located in Silicon Valley, USA.
• Recording projects (including audio versions, transcripts, word timings and candidate clips): kept for 3 days after the last edit (7 days for accounts that have bought or redeemed credits), then permanently deleted automatically (checked every hour); a project you delete in "My recordings" is permanently deleted at once — there is no Trash. Exported files are a download cache kept for at most 2 hours. Intermediate temporary files are usually cleaned up within 1 hour, reference clips and recordings for Remove fillers uploaded to Alibaba Cloud Model Studio's temporary storage are deleted automatically after at most 48 hours, and the temporary voices registered from them are deleted as described in Section 1.3.
• Server backups: once a day, containing only the account database, audit and consent records and the text of recording projects (names, transcripts and the like) — no recording audio; backups older than 7 days are deleted automatically (at most 14 are kept). Deleted data (including data of deleted accounts) disappears from backups within 7 days as they rotate.
• Compliance audit logs and voice-consent records: kept for at least 6 months and deleted after 3 years by default. When you delete your account, consent records have their personal fields removed as described in Section 6 and are deleted once 6 months have passed since the consent.
• Account information: kept until you delete your account and deleted immediately when you do; a sign-in on a device ends after 7 days without use (and after at most 60 days of continuous use), or when you sign out.
• Sign in with Apple refresh token: kept with the account; when you delete your account we immediately ask Apple to revoke the authorization and delete the token once that succeeds; if Apple cannot process it at the time, we try up to 4 more times within about 7 hours and then delete it.
• Credit balance, credit lots, usage metering and redemption records: kept until you delete your account. Credit history and purchase orders (including in-app purchase transactions): kept until you delete your account; afterwards they are kept in de-identified form (the account ID replaced by an irreversible hash, and project IDs, generation IDs and notes removed from the credit history) for 3 years for financial records, then deleted. Orders that were never paid are deleted after 30 days.

5. Security measures

• HTTPS across the site; session cookies are HttpOnly and SameSite=Strict; a strict Content Security Policy (CSP).
• Recordings and transcripts are isolated per account, and every API checks ownership; passwords are stored as salted hashes; Sign in with Apple refresh tokens are stored encrypted, with a key that is not part of data backups.
• Data files and audit logs are readable and writable only by the server process (permissions 0600).
• Sign-in, sign-up and similar APIs are rate-limited; logs do not record transcripts, audio or keys.

If a personal-information security incident occurs, we will notify you promptly and report it to the authorities as required by law.

6. Your rights

• Access and copy: you can see all your projects and transcripts in "My recordings" and export the audio at any time; for a copy of your account information or consent records, contact us at the email below.
• Correction: you can correct transcripts and change your display name and password at any time.
• Deletion: you can delete a single project in "My recordings" at any time; it is permanently deleted at once and cannot be restored. Projects you do not delete are permanently deleted automatically 3 days after their last edit (7 days for accounts that have bought or redeemed credits). Deleting your account immediately deletes all projects.
• Withdrawing consent: you can withdraw consent to voice processing by deleting the recording project concerned or deleting your account; withdrawal does not affect processing already carried out on the basis of consent.
• Deleting your account: you can delete your account yourself: after signing in, open the account menu (bottom left on desktop, the avatar at the top right on mobile), open "Account settings", choose "Delete account" and confirm by typing the phrase shown (email-and-password accounts also enter the current password). Deletion cannot be undone; export any recordings you want to keep first.
  • Deleted immediately: account information (email, display name, password hash, agreement records), third-party sign-in links (for Sign in with Apple, we also ask Apple to revoke your authorization for this service; see Section 4), all sessions (every device is signed out at once), all recording projects (original audio, audio versions, replacement clips, exported files, transcripts and transcription progress); the credit balance and credit lots (remaining credits lapse with the account), generation counts and transcription usage, the link between redeem codes and the account, and the account's random identifier for in-app purchases. Any transcription or generation in progress is stopped first.
  • Kept as required by law: voice-consent records keep only the consent ID, account ID, consent type, project ID, consent-text version and confirmation time; the IP address and browser information are deleted at account deletion, and the record is deleted automatically 6 months after the consent. Compliance audit logs are kept for the periods in Section 4 (at least 6 months, 3 years by default) and then deleted; the deletion event records only the account ID, time and number of items deleted, and log entries recorded before deletion (which may include IP address and browser information) are not rewritten and are deleted when they expire. Credit history is kept in de-identified form for 3 years for financial records. Copies in server backups are deleted within 7 days as backups rotate.
  • After deletion you can sign up again with the same email or third-party account; you get a brand-new empty account and cannot recover the old data.
  • The legacy shared workspace (access passphrase) is shared by several people and cannot be deleted; its recordings can be deleted in "My recordings". If you cannot sign in, you can also email lynxzhang1988@gmail.com to request deletion; we handle it within 15 business days after verifying your identity.
• We normally reply to the requests above within 15 business days.

7. Minors

The Service is intended mainly for adults. Minors under 14 may use it only with the consent and guidance of a parent or other guardian, who should read this policy. We do not knowingly collect the voices of minors for original-voice cloning; if we find that personal information of a minor under 14 has been processed without a guardian's consent, we delete it as soon as possible.

8. Labelling of AI-generated content

When a recording containing AI-synthesized segments is exported, an AI-generated label (including the service provider's name and content ID) is written into the file metadata; exports from the Chinese interface also get a "short-long-short-short" audio cue at the end of the file by default (not by default in the English interface; either can be changed at export). The editor also marks AI-synthesized sentences. See Section 6 of the Terms of Use for details.

9. Updates to this policy

When this policy is updated, we update the version number and effective date; for material changes, you will need to read and agree again the next time you use the Service.

10. Contact us

Operator: xiang zhang
Personal-information contact email: lynxzhang1988@gmail.com

If you are not satisfied with our reply, you can complain to or report to the competent authorities, such as the cyberspace administration and public security authorities.