AcousticCorpusMixin#

class montreal_forced_aligner.corpus.acoustic_corpus.AcousticCorpusMixin(audio_directory=None, **kwargs)[source]#

Bases: CorpusMixin, FeatureConfigMixin

Mixin class for acoustic corpora

Parameters:: audio_directory (str) – Extra directory to look for audio files

See also

CorpusMixin: For corpus parsing parameters
FeatureConfigMixin: For feature generation parameters

Variables:

sound_file_errors (list[str]) – List of sound files with errors in loading
stopped (Event) – Stop check for loading the corpus

calc_cmvn()[source]#

Calculate CMVN statistics for speakers

See also

featbin/compute-cmvn-stats.cc: Relevant Kaldi binary

calc_fmllr(iteration=None)[source]#

Multiprocessing function that computes speaker adaptation transforms via feature-space Maximum Likelihood Linear Regression (fMLLR).

See also

CalcFmllrFunction: Multiprocessing helper function for each job
AcousticCorpusMixin.calc_fmllr_arguments: Job method for generating arguments for the helper function
align_fmllr.sh: Reference Kaldi script
train_sat.sh: Reference Kaldi script

calc_fmllr_arguments(iteration=None)[source]#

Generate Job arguments for CalcFmllrFunction

Returns:: Arguments for processing
Return type:: list[CalcFmllrArguments]

combine_feats()[source]#: Combine feature generation results and store relevant information

compute_vad()[source]#

Compute Voice Activity Detection features over the corpus

See also

ComputeVadFunction: Multiprocessing helper function for each job
AcousticCorpusMixin.compute_vad_arguments: Job method for generating arguments for helper function

compute_vad_arguments()[source]#

Generate Job arguments for ComputeVadFunction

Returns:: Arguments for processing
Return type:: list[VadArguments]

create_corpus_split()[source]#: Create the split directory for the corpus

final_feature_arguments()[source]#

Generate Job arguments for MfccFunction

Returns:: Arguments for processing
Return type:: list[MfccArguments]

generate_features()[source]#: Generate features for the corpus

generate_final_features()[source]#: Generate features for the corpus

get_feat_dim()[source]#

Calculate the feature dimension for the corpus

Returns:: Dimension of feature vectors
Return type:: int

inspect_database()[source]#: Check if a database file exists and create the necessary metadata

load_corpus()[source]#: Load the corpus

mfcc()[source]#

Multiprocessing function that converts sound files into MFCCs.

See Kaldi feat page for an overview on feature generation in Kaldi.

See also

MfccFunction: Multiprocessing helper function for each job
AcousticCorpusMixin.mfcc_arguments: Job method for generating arguments for helper function
make_mfcc.sh: Reference Kaldi script

mfcc_arguments()[source]#

Generate Job arguments for MfccFunction

Returns:: Arguments for processing
Return type:: list[MfccArguments]

property no_transcription_files#: List of sound files without text files

property transcriptions_without_wavs#: List of text files without sound files

validate_corpus()[source]#: Validate the loaded files