vllm_omni.protocol.realtime.formats ¶

Realtime audio-format negotiation: what a client may declare and how it is read.

A Realtime client declares audio formats in three places --- the session object (session.audio.input.format and the older flat input_audio_format), a response.create override, and a conversation.item.create audio part --- and may spell one format several ways (audio/pcm, pcm16, s16le). This module is the single place that normalizes those spellings and says which are supported. Folding a session's declaration into the per-connection append defaults is the job of session.py, which builds on this module.

Everything here is a pure function of wire payloads: no session state, no model, no engine.

REALTIME_INPUT_AUDIO_FORMATS = {
    "pcm16",
    "pcm_s16le",
    "s16le",
    "pcm_f32le",
    "g711_ulaw",
    "g711_alaw",
}

REALTIME_OUTPUT_AUDIO_FORMATS = {
    "pcm16",
    "pcm_s16le",
    "s16le",
    "wav",
    "pcm",
    "g711_ulaw",
    "g711_alaw",
}

is_supported_realtime_input_format(fmt: object) -> bool

parse_realtime_audio_format(
    raw_format: object,
) -> tuple[object, int | None]

realtime_audio_format_object(
    fmt: object, *, sample_rate_hz: int | None = None
) -> dict[str, object]

realtime_output_format(duplex_format: object) -> str

validate_conversation_item_audio_formats(
    item: object,
) -> str | None

validate_realtime_response_audio_formats(
    response_payload: Mapping[str, object],
) -> str | None

validate_realtime_session_audio_formats(
    session_payload: Mapping[str, object],
    *,
    input_audio_formats: Collection[str] | None = None,
    output_audio_formats: Collection[str] | None = None,
) -> str | None