IvectorTrainer#

class montreal_forced_aligner.ivector.trainer.IvectorTrainer(num_iterations=10, subsample=5, gaussian_min_count=100, **kwargs)[source]#

Bases: IvectorModelTrainingMixin, IvectorConfigMixin

Trainer for a block of ivector extractor training

Parameters:

num_iterations (int) – Number of iterations, defaults to 10
subsample (int) – Subsample factor for feature frames, defaults to 5
gaussian_min_count (int)

See also

IvectorModelTrainingMixin: For base parameters for ivector training
IvectorConfigMixin: For parameters for ivector feature generation

acc_ivector_stats()[source]#

Multiprocessing function that accumulates ivector extraction stats.

See also

AccIvectorStatsFunction: Multiprocessing helper function for each job
IvectorTrainer.acc_ivector_stats_arguments: Job method for generating arguments for the helper function
ivectorbin/ivector-extractor-sum-accs.cc: Relevant Kaldi binary
ivectorbin/ivector-extractor-est.cc: Relevant Kaldi binary
sid/train_ivector_extractor.sh: Reference Kaldi script

acc_ivector_stats_arguments()[source]#

Generate Job arguments for AccIvectorStatsFunction

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

property dubm_path#: DUBM model path

property exported_model_path#: Temporary directory path that trainer will save ivector extractor model

finalize_training()[source]#: Finalize ivector extractor training

gauss_to_post()[source]#

Multiprocessing function that does Gaussian selection and posterior extraction

See also

GaussToPostFunction: Multiprocessing helper function for each job
IvectorTrainer.gauss_to_post_arguments: Job method for generating arguments for the helper function
sid/train_ivector_extractor.sh: Reference Kaldi script

gauss_to_post_arguments()[source]#

Generate Job arguments for GaussToPostFunction

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

property ie_path#: Current ivector extractor model path

property ivector_options#: Options for ivector training and extracting

property meta#: Metadata information for ivector extractor models

property next_ie_path#: Next iteration’s ivector extractor model path

train_iteration()[source]#: Run an iteration of training

property train_type#: Training identifier