Voice Chat

UseBelha supports real-time voice conversations through a secure WebSocket proxy (/api/live). Users can speak naturally and hear responses instantly.

Emotion-Aware Voice

Voice templates can enable emotion-aware responses for more natural conversations.

Voice Templates

Voice Assistant

tpl_voice

Friendly voice assistant for general conversations. Uses Puck voice.

Voice Support Agent

tpl_voice_support

Empathetic customer support via voice. Uses Kore voice.

Available Voices

Voice	Description	Style
`Puck`	Friendly and casual	Neutral
`Charon`	Professional and calm	Male
`Kore`	Warm and empathetic	Female
`Fenrir`	Confident and energetic	Male
`Aoede`	Clear and articulate	Female

Voice Configuration

Add voice support to any agent:

json

{
  "name": "My Voice Agent",
  "systemPrompt": "You are a helpful voice assistant...",
  "voice": {
    "enabled": true,
    "voiceName": "Kore",
    "affectiveDialog": true
  }
}

How It Works

Connection - Browser connects to /api/live WebSocket proxy
Audio Capture - Microphone captures 16-bit PCM at 16kHz
Streaming - Audio streams through the backend proxy to the voice service
Response - AI generates audio response in real-time
Playback - Response plays through speakers at 24kHz

Using the useLiveChat Hook

typescript

import { useLiveChat } from '@/lib/useLiveChat';

const {
  isConnected,      // WebSocket connected
  isConnecting,     // Connection in progress
  isListening,      // Microphone active
  isSpeaking,       // AI generating audio
  messages,         // Chat history with transcriptions
  connect,          // Start connection
  disconnect,       // Close connection
  startListening,   // Begin voice capture
  stopListening,    // Stop voice capture
  sendText,         // Send text message
  interrupt,        // Stop current playback
} = useLiveChat({
  systemPrompt: 'You are a helpful assistant.',
  voiceName: 'Kore',
  affectiveDialog: true,
});

Audio Requirements

Direction	Format	Sample Rate
Input (microphone)	16-bit PCM, little-endian	16kHz
Output (speakers)	16-bit PCM, little-endian	24kHz

Testing Voice

Open the Dashboard
Select a voice template (tpl_voice or tpl_voice_support)
Click Quick Test
Allow microphone access when prompted
Click the microphone button to start speaking

Browser Support

Voice requires a modern browser with WebSocket, AudioWorklet, and microphone permissions. Chrome, Edge, and Firefox are recommended.