MonophoneTrainer

class montreal_forced_aligner.acoustic_modeling.MonophoneTrainer(subset=2000, initial_gaussians=135, initial_beam=6, max_gaussians=1000, power=0.25, boost_silence=1.25, **kwargs)

Bases: AcousticModelTrainingMixin

Configuration class for monophone training

Variables:

• subset (int) – Number of utterances to use, defaults to 2000

• initial_gaussians (int) – Number of gaussians to begin training, defaults to 135

• max_gaussians (int) – Total number of gaussians, defaults to 1000

• power (float) – Exponent for number of gaussians according to occurrence counts, defaults to 0.25

See also

AcousticModelTrainingMixin

For acoustic model training parsing parameters

property align_options

Alignment parameters

compute_calculated_properties()

Generate realignment iterations and initial gaussians based on configuration

mono_align_equal()

Multiprocessing function that creates equal alignments for base monophone training.

See also

MonoAlignEqualFunction

Multiprocessing helper function for each job

MonophoneTrainer.mono_align_equal_arguments

Job method for generating arguments for the helper function

gmmbin/gmm-sum-accs.cc

Relevant Kaldi binary

gmmbin/gmm-est.cc

Relevant Kaldi binary

train_mono.sh

Reference Kaldi script

mono_align_equal_arguments()

Generate Job arguments for MonoAlignEqualFunction

Returns:

Arguments for processing

Return type:

list[MonoAlignEqualArguments]

property phone_type

Phone type

property train_type

Training identifier