Skip to content

vllm.entrypoints.openai.realtime.api_router

logger module-attribute 

logger = init_logger(__name__)

router module-attribute 

router = APIRouter()

attach_router 

attach_router(app: FastAPI)

Attach the realtime router to the FastAPI app.

Source code in vllm/entrypoints/openai/realtime/api_router.py 
53
54
55
56
def attach_router(app: FastAPI):
    """Attach the realtime router to the FastAPI app."""
    app.include_router(router)
    logger.info("Realtime API router attached")

init_realtime_state 

init_realtime_state(
    engine_client: EngineClient,
    state: State,
    args: Namespace,
    request_logger: RequestLogger | None,
    supported_tasks: tuple[SupportedTask, ...],
)
Source code in vllm/entrypoints/openai/realtime/api_router.py 
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
def init_realtime_state(
    engine_client: "EngineClient",
    state: "State",
    args: "Namespace",
    request_logger: RequestLogger | None,
    supported_tasks: tuple["SupportedTask", ...],
):
    state.openai_serving_realtime = (
        OpenAIServingRealtime(
            engine_client,
            state.openai_serving_models,
            request_logger=request_logger,
            log_error_stack=args.log_error_stack,
        )
        if "realtime" in supported_tasks
        else None
    )

realtime_endpoint async 

realtime_endpoint(websocket: WebSocket)

WebSocket endpoint for realtime audio transcription.

Protocol: 1. Client connects to ws://host/v1/realtime 2. Server sends session.created event 3. Client optionally sends session.update with model/params 4. Client sends input_audio_buffer.commit when ready 5. Client sends input_audio_buffer.append events with base64 PCM16 chunks 6. Server processes and sends transcription.delta events 7. Server sends transcription.done with final text + usage 8. Repeat from step 5 for next utterance 9. Optionally, client sends input_audio_buffer.commit with final=True to signal audio input is finished. Useful when streaming audio files

Audio format: PCM16, 16kHz, mono, base64-encoded

Source code in vllm/entrypoints/openai/realtime/api_router.py 
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
@router.websocket("/v1/realtime")
async def realtime_endpoint(websocket: WebSocket):
    """WebSocket endpoint for realtime audio transcription.

    Protocol:
    1. Client connects to ws://host/v1/realtime
    2. Server sends session.created event
    3. Client optionally sends session.update with model/params
    4. Client sends input_audio_buffer.commit when ready
    5. Client sends input_audio_buffer.append events with base64 PCM16 chunks
    6. Server processes and sends transcription.delta events
    7. Server sends transcription.done with final text + usage
    8. Repeat from step 5 for next utterance
    9. Optionally, client sends input_audio_buffer.commit with final=True
       to signal audio input is finished. Useful when streaming audio files

    Audio format: PCM16, 16kHz, mono, base64-encoded
    """
    app = websocket.app
    serving = app.state.openai_serving_realtime

    connection = RealtimeConnection(websocket, serving)
    await connection.handle_connection()