VoiceAgent

VoiceAgent is the main entry point. It manages microphone capture (AudioWorklet), scheduled PCM playback, and session lifecycle on top of any Transport.

---

Factory methods

VoiceAgent.withWebSocket(config, options)

const agent = VoiceAgent.withWebSocket(
  { url: 'ws://localhost:8000/ws' },
  { onStatus, onMessage, onError }
);

VoiceAgent.withWebRTC(config, options)

const agent = VoiceAgent.withWebRTC(
  { baseUrl: 'http://localhost:8000', metadata: { session_type: 'default' } },
  { onStatus }
);

new VoiceAgent(transport, options)

Pass any Transport — including a TransportSwitcher — directly.

---

Options

Option	Type	Description
onStatus	(status: SessionStatus) => void	idle → connecting → connected → listening → speaking → ended
onMessage	(msg: BotMessage) => void	JSON control messages from the server
onQoS	(snap: QoSSnapshot) => void	Live quality metrics (requires QoSMonitor)
onError	(err: Error) => void	Transport-level errors
onLog	(msg: string) => void	Internal diagnostic log

---

Methods

Method	Description
connect()	Opens transport + requests mic permission
disconnect()	Closes transport + releases mic
setMicEnabled(enabled)	Mute / unmute microphone
sendMessage(msg)	Send a JSON control message to the server

---

Audio path (WebSocket)

getUserMedia (16 kHz, mono, AEC+NS)
    └── AudioWorklet "pcm-sender"
            ├── Int16Array conversion
            └── sendAudio(buffer) → WebSocket binary frame

WebSocket binary frame (PCM16 from server)
    └── Int16Array → Float32Array
            └── AudioBufferSourceNode (scheduled, gapless)
                    └── MediaStreamDestinationNode (AEC reference)