Skip to main content

Speech

Speech Panel

The Speech panel provides speech-to-text (STT), text-to-speech (TTS), voice cloning, and full voice conversation capabilities. These features require the speech addons to be installed (see Addons).

Speech-to-Text (STT)​

Record audio and get transcriptions:

  • Model selection — choose from available Whisper models (base, small, medium, large)
  • Language — set the source language for better accuracy
  • Word timestamps — optionally get per-word timing data
  • Live waveform — visualize audio input in real-time

Recording​

Click the microphone button to start recording. The waveform visualization shows audio levels in real-time. Click stop when done, and the transcription appears below.

If microphone access hasn't been granted, a permission prompt guides you through enabling it.

Text-to-Speech (TTS)​

Convert text to spoken audio:

  • Model selection — choose from available TTS models (e.g., pocket-tts)
  • Voice selection — pick from available voices for the selected model
  • Audio playback — listen to generated audio with built-in player
  • History — previous TTS generations are kept for reference

Voice Cloning​

Create custom voices from audio samples:

  • Record or upload a voice sample
  • Clone the voice for use with TTS
  • Use cloned voices in conversation mode

Conversation Mode​

Full voice-to-voice conversation with your AI:

  • Turn detection — configurable settings for when the AI should start responding
  • Continuous conversation — alternating speech turns with the AI
  • Conversation view — see the full transcript with speaker labels
  • Model integration — uses your project's configured inference model for responses

Turn Detection Settings​

Configure how the system detects when you've stopped talking:

  • Silence threshold
  • Minimum speech duration
  • Response delay

Requirements​

Speech features require the Universal Runtime (port 11540) to be running. The panel checks runtime health automatically and shows status badges for STT/TTS availability.

If addons aren't installed, an inline prompt lets you install them directly or navigate to the Addons page.

API Routes​

ActionMethodRoute
Transcribe audioPOST/v1/audio/transcriptions (Universal Runtime)
Synthesize speechPOST/v1/{ns}/{project}/audio/speech
List voicesGET/v1/{ns}/{project}/audio/voices
Stream transcriptionWebSocket/v1/audio/transcriptions/stream (Universal Runtime)

Route​

Speech is accessed through the Test page in speech mode:

/chat/test (select Speech mode)