vllm.entrypoints.speech_to_text.realtime.protocol ¶

class ErrorEvent(OpenAIBaseModel):
    """Error notification."""

    type: Literal["error"] = "error"
    error: str
    code: str | None = None

class InputAudioBufferAppend(OpenAIBaseModel):
    """Append audio chunk to buffer."""

    type: Literal["input_audio_buffer.append"] = "input_audio_buffer.append"
    audio: str  # base64-encoded PCM16 @ 16kHz

class InputAudioBufferCommit(OpenAIBaseModel):
    """Process accumulated audio buffer."""

    type: Literal["input_audio_buffer.commit"] = "input_audio_buffer.commit"
    final: bool = False

class SessionCreated(OpenAIBaseModel):
    """Connection established notification."""

    type: Literal["session.created"] = "session.created"
    id: str = Field(default_factory=lambda: f"sess-{random_uuid()}")
    created: int = Field(default_factory=lambda: int(time.time()))

class SessionUpdate(OpenAIBaseModel):
    """Configure session parameters."""

    type: Literal["session.update"] = "session.update"
    model: str | None = None

class TranscriptionDelta(OpenAIBaseModel):
    """Incremental transcription text."""

    type: Literal["transcription.delta"] = "transcription.delta"
    delta: str  # Incremental text

class TranscriptionDone(OpenAIBaseModel):
    """Final transcription with usage stats."""

    type: Literal["transcription.done"] = "transcription.done"
    text: str  # Complete transcription
    usage: UsageInfo | None = None