← Back to API reference

neuralset.extractors.audio.Whisper

pydantic model neuralset.extractors.audio.Whisper

Extract speech embeddings using the pretrained Whisper model from Hugging Face.

Whisper is a multilingual speech recognition and translation model that includes a dedicated encoder for audio processing. This class provides an interface to convert raw audio waveforms into high-level embeddings suitable for automatic speech recognition (ASR), speech translation, and other downstream tasks.

model_name

The Hugging Face model identifier to load. Defaults to "openai/whisper-large-v3-turbo".

Type:

str

Fields:

• dtype (Literal['float32'])

• hf_config (neuralset.extractors.audio.HuggingFaceAudioConfig)

• model_name (str)

field model_name: str = 'openai/whisper-large-v3-turbo'

field dtype: Literal['float32'] = 'float32'

field hf_config: HuggingFaceAudioConfig = HuggingFaceAudioConfig(** { 'model_cls_name': 'WhisperModel',   'model_kwargs': None,   'processor_cls_name': 'AutoFeatureExtractor',   'processor_kwargs': None} )

load_model() → Module

requirements: tp.ClassVar[tuple[str, ...]] = ('transformers>=4.29.2', 'huggingface_hub>=0.27.0', 'julius>=0.2.7', 'pillow>=9.2.0', 'transformers>=4.29.2', 'huggingface_hub>=0.27.0', 'julius>=0.2.7', 'pillow>=9.2.0', 'transformers>=4.29.2', 'soundfile', 'transformers>=4.29.2', 'huggingface_hub>=0.27.0', 'julius>=0.2.7', 'pillow>=9.2.0', 'transformers>=4.29.2', 'huggingface_hub>=0.27.0', 'julius>=0.2.7', 'pillow>=9.2.0', 'transformers>=4.29.2', 'soundfile', 'soundfile')

← Back to API reference