Realtime Streaming (/v1/realtime)

The /v1/realtime endpoint provides standard full-duplex WebSocket audio streaming for live captions and interactive voice agents.

---

Connection Endpoint

ws://localhost:8000/v1/realtime?api_key=sk-oneasr-v1-xxx&model=xasr

---

Audio Input Format

• Sample Rate: 16,000 Hz (16 kHz)
• Channels: 1 (Mono)
• Encoding: 16-bit Signed Integer PCM Little-Endian (pcm_s16le)
• Chunk Duration: 100ms - 200ms per frame (3,200 bytes per chunk)

---

Message Protocol

Client ➔ Server:

Stream raw binary PCM audio bytes continuously over the WebSocket connection.

Server ➔ Client (JSON Text):

{
  "type": "transcription",
  "text": "Hello world",
  "is_final": false,
  "start": 0.0,
  "end": 1.2
}

When an endpoint sentence boundary is detected:

{
  "type": "transcription",
  "text": "Hello world.",
  "is_final": true,
  "start": 0.0,
  "end": 1.8
}