Privacy policy

Scope

This policy applies to the macOS edition of SpeechSub. SpeechSub performs speech recognition and caption proofreading on your device. You do not need a SpeechSub account. The current product has no payment, account, or telemetry service, and no API that sends audio, video, or captions to the cloud for recognition.

Files processed on your device

Audio, video, speech recognition, and caption editing run on your device. The current version does not upload the media files or caption content you choose to SpeechSub’s operator servers, and does not send them as recognition content to a developer-operated recognition service. The app reads files you select and stores caption results, drafts, glossaries, recognition history, and necessary temporary files in the local user data directory.

Data stored locally

Local data may include references to media paths you chose, temporary converted audio, caption results, glossaries you create or import, named drafts, recognition history, interface settings, downloaded model files, and local logs. The voice-activity file ggml-silero-v6.2.0.bin (885,098 bytes), used only to distinguish speech from silence, ships inside the app bundle; it is not a recognition model you are asked to download separately. This data stays on your computer and is not uploaded to SpeechSub’s operator servers. You can delete drafts, clean temporary recognition files, or remove an individual history record in the app. Those actions do not delete your original media. Before uninstalling or deleting files manually, back up anything you still need.

Model download and network use

Recognition runs locally. If a required Whisper model is not yet present, the app downloads that tier on demand: Turbo (about 833.7 MB, ggml-large-v3-turbo-q8_0.bin), Small (about 252.2 MB, ggml-small-q8_0.bin), or Base (about 78.0 MB, ggml-base-q8_0.bin). The download uses the official source huggingface.co/ggerganov/whisper.cpp first, fails over to the backup line hf-mirror.com, and verifies the file with a SHA-256 checksum. Model files are stored in the user data directory, not in the app bundle.

Optional timing-correction model

If you choose “Recognize and correct timing · Whisper + Qwen,” the app may download on demand a model of about 1.8 GB used only to correct caption timing, not to transcribe. The download should show the size and ask you to confirm, use the official source first, verify the file, and store it in the user data directory. If that model has not been downloaded, the mode should prompt you to download it and should not finish silently.

Technical information that download servers may process

When the app requests model files from the official source or the backup line, those servers may process technical connection data, such as IP addresses, under their own policies. This is not an upload of your media or captions as recognition content. SpeechSub has no cloud recognition API. The current product has no account, payment, or telemetry service. If that changes, an updated policy will explain it.

Permissions

The app should request only the file access needed for actions you initiate, such as reading media you pick with the system file picker.

If distributed on the Mac App Store

If the app is distributed through the Mac App Store, Apple handles the purchase process. The current product has no developer-operated account or payment system, and does not receive your Apple ID password or payment card number. Apple’s services are also subject to Apple’s privacy policy.

Contact, support records, and rights requests

For support or privacy requests, email 6666699999@live.com. If you contact us, the contact details, problem description, and attachments you provide enter the support channel you chose. Please do not send passwords, keys, complete media files, or unredacted transcripts. Support records are kept under a 90-day retention practice after the issue is resolved, except where a legal obligation or an unresolved dispute requires longer retention. They are managed manually by the developer; there is no automatic deletion, and this does not guarantee that an email provider’s backups have been completely removed. You may use the same address to request access, correction, or deletion, and such requests are handled under applicable law.

Public pages and website access

This site is at speechsub.ahzjjs.com. Legal pages may also be hosted separately. The hosting platform may log visitor IP addresses and other access data for security and operations. This site does not use advertising or third-party analytics scripts, does not collect form data, and does not maintain a user database. The hosting platform still processes access data under its rules, and email providers may process messages and backups under theirs.

Updates

This policy will be updated when product behavior or operator details are confirmed. Material changes should be highlighted in the product or on the website. Changes that add new data sending or new purposes must be explained before use, and any legally required authorization must be obtained.