Local processing and privacy

Where audio is recognizedCopy Anchor Link

• Local recognition: After downloading and configuring a model, audio is processed on the device and recognition can work offline.
• Online recognition: Recorded audio is sent to your selected provider. Its policies govern retention and use.

Where polishing text goesCopy Anchor Link

• Online language model: Translation or polishing sends the text and template instructions to the selected service. Hotword enhancement may also include hotwords.
• Local language model: A service such as Ollama running on your device processes the text locally. Check that the configured endpoint really points to your device.

Local recognition does not automatically make later polishing local.

Keep audio and text processing localCopy Anchor Link

1. Download and enable a local recognition model under Audio Models.
2. Use plain transcription without online polishing, or configure a language model running on your device.
3. Once downloads finish, verify the configuration with an actual input.

Network activity for model downloads or update checks is separate from where audio and text are processed.

Local history and recordingsCopy Anchor Link

The app provides local history for reviewing results and retrying failed inputs. App Settings → Keep Recordings controls whether recordings from the most recent month are retained. Failed recordings are still retained for retry when this option is off.

To control retention on your device, review both history and recording settings. Local processing does not mean nothing is saved locally.

Fees and accountsCopy Anchor Link

VoicePaste is free software. Online recognition and polishing providers may charge separately. API keys are your service credentials; do not share them or include them in public screenshots.