vllm_omni.entrypoints.duplex.audio_encoding ¶

Audio encoding for duplex model output (entrypoint layer, injected into DuplexOmniEngine).

logger `module-attribute` ¶

logger = init_logger(__name__)

encode_audio(
    audio_data: object,
    sample_rate_hz: int,
    response_format: str,
    speed: float | None,
) -> str | None

Encode a model audio tensor/array into base64 in response_format.

Moved from the serving runtime bridge; uses the shared AudioMixin encoder directly instead of going through the chat service.