asr.hf.transcribe

asr.hf.transcribe(
    model,
    processor,
    file_dataloader,
    language=None,
    task='transcribe',
    batch_size=4,
    beam_size=3,
    length_penalty=1.0,
    repetition_penalty=1.0,
    max_length=250,
    num_workers=2,
    prefetch_factor=2,
    output_dir='output/transcriptions',
    device='cuda',
)

Transcribe audio files using HuggingFace Whisper model.

Parameters

Name	Type	Description	Default
model	transformers.WhisperForConditionalGeneration	HuggingFace Whisper model.	required
processor	transformers.WhisperProcessor	HuggingFace Whisper processor.	required
file_dataloader	torch.utils.data.DataLoader	DataLoader yielding audio file datasets.	required
language	str	Language code (e.g., ‘sv’, ‘en’). Default is None (auto-detect).	None
batch_size	int	Batch size for inference.	4
beam_size	int	Number of beams for beam search. Default is 3.	3
length_penalty	float	Length penalty. Default is 1.0.	1.0
repetition_penalty	float	Repetition penalty. Default is 1.0.	1.0
max_length	int	Maximum length of generated text. Default is 250.	250
num_workers	int	Number of workers for feature dataloader.	2
prefetch_factor	int	Prefetch factor for feature dataloader.	2
output_dir	str	Directory to save transcription JSON files. Default is output/transcriptions.	'output/transcriptions'
device	str	Device to run inference on. Default is cuda.	'cuda'