nemo_curator.models.asr.faster_whisper

Faster-Whisper implementation of the shared ASR adapter.

Module Contents

Classes

Name	Description
`FasterWhisperASR`	Run Faster-Whisper on mono 16 kHz waveforms prepared by `ASRStage`.

Functions

Name	Description
`_download_whisper_model`	-
`_faster_whisper_stack`	-
`_whisper_model_class`	-

Data

_LANGUAGE_ALIASES

_TARGET_SAMPLE_RATE

API

class nemo_curator.models.asr.faster_whisper.FasterWhisperASR(
    model_id: str = 'large-v3',
    revision: str | None = None,
    compute_type: str = 'float16',
    beam_size: int = 5,
    vad_filter: bool = True,
    without_timestamps: bool = True,
    cpu_compute_type: str = 'int8'
)

Dataclass

Run Faster-Whisper on mono 16 kHz waveforms prepared by ASRStage.

_model

Any | None = field(default=None, init=False, repr=False)

beam_size

int = 5

compute_type

str = 'float16'

cpu_compute_type

str = 'int8'

model_id

str = 'large-v3'

revision

str | None = None

vad_filter

bool = True

without_timestamps

bool = True

nemo_curator.models.asr.faster_whisper.FasterWhisperASR.__post_init__() -> None

nemo_curator.models.asr.faster_whisper.FasterWhisperASR.download_weights_on_node() -> None

Populate Faster-Whisper’s node-local cache without allocating a GPU.

nemo_curator.models.asr.faster_whisper.FasterWhisperASR.load_model(
    num_gpus: int
) -> None

nemo_curator.models.asr.faster_whisper.FasterWhisperASR.transcribe_batch(
    items: list[dict[str, typing.Any]]
) -> list[nemo_curator.models.asr.base.ASRResult]

nemo_curator.models.asr.faster_whisper.FasterWhisperASR.unload_model() -> None

nemo_curator.models.asr.faster_whisper._download_whisper_model(
    model_id: str,
    revision: str | None
) -> None

nemo_curator.models.asr.faster_whisper._faster_whisper_stack() -> tuple[type, typing.Any]

nemo_curator.models.asr.faster_whisper._whisper_model_class() -> type

nemo_curator.models.asr.faster_whisper._LANGUAGE_ALIASES = {'fil': 'tl', 'in': 'id', 'iw': 'he', 'ji': 'yi', 'jv': 'jw', 'nb': 'no', 'tl': ...

nemo_curator.models.asr.faster_whisper._TARGET_SAMPLE_RATE = 16000