Our approach
Meizin Voice performs speech recognition on your PC when you use the default Whisper engine. From v1.6, you can optionally enable Soniox, which sends audio to Soniox for recognition. License-related communication is separate from your choice of recognition engine.
Audio and recognition results
From version 1.8.0, stopping a recording saves the original audio as a WAV file on your PC in AppData\Roaming\meizin-voice\recordings, whether recognition succeeds or fails. These files are not automatically deleted by age or count. Recognized text is inserted into the destination and saved in recognition history on your PC. History and diagnostic logs may contain recognized text. The developer does not automatically collect your audio or recognition history.
Optional Soniox processing
Soniox is an optional external service. If you save your own API key and enable it, audio from your selected input source, recognition settings such as language, and your configured terms and background context are sent to Soniox over an encrypted connection. The current app identifies its endpoint as United States. Turning Soniox off returns subsequent recognition to Whisper.
Soniox processing, retention, and usage records are governed by Soniox's privacy policy and your agreement with Soniox. API charges are separate from the Meizin Voice license. Retrieving overall Soniox usage makes an authenticated request to the Soniox usage API.
Optional text conversion
Turning Conversion instructions ON sends the Soniox transcript and your saved instructions to OpenAI through your separately configured Codex. Recorded audio is not sent to OpenAI for this text conversion. This optional feature requires separately configured WSL Ubuntu-24.04, Python 3, and an authenticated Codex CLI. OpenAI processing is subject to OpenAI's privacy policy and your agreement. Turning it OFF skips conversion. Normal local Whisper recognition does not require Codex or WSL.
Data stored on your PC
- Display language, hotkey, input source, model, window position, and other settings
- Correction dictionary, recognition history, original WAV recordings, conversion instructions, and diagnostic logs
- Input audio duration, monthly totals, and cumulative totals for Whisper and Soniox. These remain when recognition history is deleted
- Soniox settings, recognition hints, and your API key encrypted using Windows data protection
- License key, activation details, trial start date, and offer information
- The bundled Small model and any additional recognition models you download
Network communication
- Soniox recognition when enabled, and requests for Soniox usage information
- Sending transcripts and instructions to OpenAI through separately configured Codex when Conversion instructions is enabled
- Downloading additional models or restoring required model files from distribution servers
- Trial offer registration, using the app identifier, machine identifier, and app version. Device identifiers are hashed on the license server
- License validation and transfer, using the license key, device identifier, version, and related required information
- Microsoft Store updates and use of help, purchase, contact, and other website links
If you submit recognized text through another application or website, that destination's terms and privacy policy also apply.
Sharing
The developer does not sell or rent your data. Data is not provided to third parties except for the optional Soniox and OpenAI processing and purchase, activation, and support operations described above, with your explicit consent, or when required by law.
Purchases and support
Purchases are processed through sales channels such as Stripe and Lemon Squeezy, under their privacy policies. The developer does not directly retain credit card details. Purchaser email, order ID, license key, and activated-device information may be retained for license issuance, purchase verification, support, and transfers.
Deleting data
Recognition history can be deleted using the app's history controls. Deleting history does not delete original WAV recordings. Exit the app and delete selected WAV files from the recordings folder above to remove them. Settings, dictionary, usage totals, and models are stored separately; deleting one configuration file or uninstalling the app may not remove all data. Contact us for help removing remaining data or for purchase and activation data requests. Manage your Soniox account and service records through Soniox.
Children
Meizin Voice is not directed at children under 13. The developer does not knowingly collect personal information from children under 13.
Changes
We may update this policy as needed. Significant changes will be announced on the website or in the app.
Contact
Use our contact form.
Last updated: October 14, 2026