Voice system
Jarvis is designed to behave like a practical desktop voice assistant: wake, listen, execute, verify, and respond naturally.
Wake words
The built-in wake phrases are "Hey Jarvis" and "Jarvis". When the wake loop detects a phrase, Jarvis either acknowledges and listens for the next command or executes the command included after the wake phrase.
Language support
The default speech languages are English and Malayalam. The UI exposes language mode so users can choose automatic detection, English, or Malayalam.
Speech recognition providers
| Provider | Use case |
|---|---|
| Google Speech Recognition | Default cloud fallback for quick local development. |
| Sarvam STT | Indic speech recognition when configured with a Sarvam key. |
| Local Whisper | Offline transcription when the Whisper package and model are installed. |
Text to speech
Edge TTS is the default voice output. Sarvam TTS can be configured for Indian language voices. Audio files are generated in backend runtime storage and ignored by Git.
Permissions
Voice activation requires `voice.listen` and background wake mode requires `voice.background`. If these are disabled, Jarvis reports that the action is blocked instead of silently listening.
Troubleshooting
- Check Windows microphone permissions.
- Confirm the backend has access to the selected audio device.
- Disable background listening in Safe Mode if you want manual control.
- If Local Whisper is selected, install the package and download a model first.
Created by Jojin John
JX Jarvis is created by Jojin John.