Izwi app iconIzwi

Voice Mode

Voice mode enables real-time spoken conversations with AI. Speak naturally and receive spoken responses — all processed locally on your device.

Overview

Voice mode combines:

  • Speech recognition — Converts your voice to text
  • AI chat — Processes your message and generates a response
  • Text-to-speech — Speaks the response back to you

Everything runs locally with no cloud services.

Getting Started

Required Models

Download the necessary models:

# Text-to-speech izwi pull Qwen3-TTS-12Hz-0.6B-Base # Speech recognition izwi pull Qwen3-ASR-0.6B-GGUF # Chat izwi pull Qwen3-8B-GGUF # Optional unified speech model izwi pull LFM2.5-Audio-1.5B-GGUF

Start Voice Mode

  1. Start the server:
    izwi serve
  2. Open the web UI:
    http://localhost:8080/voice
  3. Click the microphone button to start speaking

Using Voice Mode

Web UI

  1. Navigate to Voice in the sidebar
  2. Click the microphone button to start recording
  3. Speak your message
  4. Click again to stop recording (or wait for auto-detection)
  5. Listen to the AI response

Controls

ControlAction
MicrophoneStart/stop recording
SpeakerMute/unmute assistant responses
SettingsConfigure models, playback speed, and speech detection

Configuration

Select Voice

Choose from available voices in the settings panel. Different TTS models offer different voice options.

Select Models

Configure which models to use:

  • ASR Model — For speech recognition
  • TTS Model — For speech synthesis
  • Chat Model — For response generation

Voice Agent Prompt

Use the Voice Agent Prompt section in settings to customize the assistant's speaking style and behavior. Prompt changes are saved locally and apply the next time you start a voice session.

Observational Memory

Use Observational Memory in settings to review, enable, disable, or delete the stable user memories captured from modular voice conversations. You can forget individual memories or clear them all at any time.

Audio Settings

  • Auto-detect silence — Automatically stop recording when you stop speaking
  • Playback speed — Adjust response playback speed from 0.75x to 1.75x
  • Mute output — Silence assistant playback without ending the session

Tips for Best Results

  1. Use a good microphone — Built-in laptop mics work but external mics are better
  2. Minimize background noise — Find a quiet environment
  3. Speak clearly — Natural pace, clear pronunciation
  4. Wait for responses — Let the AI finish before speaking again

Keyboard Shortcuts

ShortcutAction
SpaceStart or stop the current voice session
EscapeStop the current voice session
MMute/unmute
Shortcuts are ignored while focus is inside a text field or other editable control.

Troubleshooting

No audio input detected

  1. Check your microphone permissions in system settings
  2. Ensure the correct input device is selected
  3. Test your microphone in another application

Responses are slow

  1. Use smaller models for faster responses
  2. Ensure models are loaded (not loading on-demand)
  3. Check system resources (RAM, CPU usage)

Poor transcription accuracy

  1. Speak more clearly and slowly
  2. Reduce background noise
  3. Try a larger ASR model (Qwen3-ASR-1.7B-GGUF)

See Also