pipeline

pipelines.pipeline(
    vad_model,
    emissions_model,
    processor,
    audio_paths,
    audio_dir,
    speeches=None,
    sample_rate=16000,
    chunk_size=30,
    alignment_strategy='speech',
    text_normalizer_fn=text_normalizer,
    tokenizer=None,
    start_wildcard=False,
    end_wildcard=False,
    blank_id=0,
    word_boundary='|',
    indent=2,
    ndigits=5,
    batch_size_files=1,
    num_workers_files=2,
    prefetch_factor_files=1,
    batch_size_features=8,
    num_workers_features=4,
    streaming=True,
    save_json=True,
    save_msgpack=False,
    save_emissions=True,
    return_alignments=False,
    delete_emissions=False,
    output_vad_dir='output/vad',
    output_emissions_dir='output/emissions',
    output_alignments_dir='output/alignments',
    device='cuda',
)

Complete pipeline to run VAD, extract emissions, and perform alignment.

Parameters

Name	Type	Description	Default
vad_model	object	The loaded VAD model.	required
emissions_model	object	The loaded ASR model.	required
processor	Wav2Vec2Processor	Wav2Vec2Processor to preprocess the audio.	required
audio_paths	list	List of paths to audio files (relative to audio_dir).	required
audio_dir	str	Base directory with audio files relative to audio_paths.	required
speeches	list[list[SpeechSegment]] or None	List of SpeechSegment objects to run VAD and alignment only on specific segments of the audio. If alignment_strategy is ‘speech’, the text needs to be supplied in the SpeechSegment objects. If alignment_strategy is ‘chunk’ and ASR transcriptions are used, there is no need to supply text in the SpeechSegment objects.	None
sample_rate	int	Sample rate to resample audio to.	16000
chunk_size	int	When alignment_strategy is set to speech, SpeechSegments are split into chunk_size sized chunks for feature extraction.	30
alignment_strategy	str	Strategy for aligning features to text. One of ‘speech’ or ‘chunk’. If speech, audio is split into chunk_size sized chunks based on SpeechSegments. If chunk, VAD chunks are used as basis for feature extraction and alignment. NOTE: chunk currently only works with ASR. The individual VAD chunks won’t contain the relevant text information for alignment.	"speech"
text_normalizer_fn	callable	Function to normalize text according to regex rules.	text_normalizer
tokenizer	object	Optional tokenizer for custom segmentation of text (e.g. sentence segmentation, or paragraph segmentation). The tokenizer should either i) be a PunktTokenizer from nltk, or ii) directly return a list of spans (start_char, end_char) when called on a string.	None
start_wildcard	bool	Whether to add a wildcard token at the start of the segments.	False
end_wildcard	bool	Whether to add a wildcard token at the end of the segments.	False
blank_id	int	ID of the blank token in the tokenizer.	0
word_boundary	str	Token indicating word boundaries in the tokenizer.	"|"
indent	int	Indentation level for saved JSON files. None to disable pretty formatting.	2
ndigits	int	Number of decimal digits to round the alignment times and scores to.	5
batch_size_files	int	Batch size for the file DataLoader.	1
num_workers_files	int	Number of workers for the file DataLoader.	2
prefetch_factor_files	int	Prefetch factor for the file DataLoader.	1
batch_size_features	int	Batch size for the feature DataLoader.	8
num_workers_features	int	Number of workers for the feature DataLoader.	4
streaming	bool	Whether to use streaming loading of audio files.	False
save_json	bool	Whether to save the output files as JSON.	True
save_msgpack	bool	Whether to save the output files as Msgpack.	False
save_emissions	bool	Whether to save the raw emissions as .npy files.	True
return_alignments	bool	Whether to return the alignment mappings.	False
delete_emissions	bool	Whether to delete the emissions files after alignment to save space.	False
output_vad_dir	str	Directory to save the VAD output files.	"output/vad"
output_emissions_dir	str	Directory to save the emissions output files.	"output/emissions"
output_alignments_dir	str	Directory to save alignment output files.	"output/alignments"
device	str	Device to run the alignment on (e.g. “cuda” or “cpu”).	"cuda"

Returns

Name	Type	Description
	list[list[SpeechSegment]] or None	If return_alignments is True, returns a list of alignment mappings for each audio file. Otherwise, returns None (the alignments are saved to disk only).