Skip to main content
Sessions created with upload_type: stream return a wss:// URL as upload_url. Connect and stream — the server auto-detects the wire format from your first frame. Option A — raw binary: send 16-bit LE, mono, 16 kHz PCM as binary frames. 
const ws = new WebSocket(uploadUrl);
ws.binaryType = "arraybuffer";
ws.onopen = () => ws.send(pcmChunk);
Option B — JSON envelope (Twilio/Vobiz Media Streams compatible), base64 PCM payloads: 
{ "event": "start", "start": { "mediaFormat": { "encoding": "audio/x-l16", "sampleRate": 16000 } } }
{ "event": "media", "media": { "payload": "<base64 PCM>" } }
{ "event": "stop" }
To finish: close the socket (flushes audio), then call End Session — closing alone does not start processing. Then poll Get Session.