LdaTrainer

class montreal_forced_aligner.acoustic_modeling.LdaTrainer(subset=10000, num_leaves=2500, max_gaussians=15000, lda_dimension=40, uses_splices=True, splice_left_context=3, splice_right_context=3, random_prune=4.0, boost_silence=1.0, power=0.25, **kwargs)

Bases: TriphoneTrainer

Triphone trainer

Parameters:

• subset (int) – Number of utterances to use, defaults to 10000

• num_leaves (int) – Number of states in the decision tree, defaults to 2500

• max_gaussians (int) – Number of gaussians in the decision tree, defaults to 15000

• lda_dimension (int) – Dimensionality of the LDA matrix

• uses_splices (bool) – Flag to use spliced and LDA calculation

• splice_left_context (int or None) – Number of frames to splice on the left for calculating LDA

• splice_right_context (int or None) – Number of frames to splice on the right for calculating LDA

• random_prune (float) – This is approximately the ratio by which we will speed up the LDA and MLLT calculations via randomized pruning

See also

TriphoneTrainer

For acoustic model training parsing parameters

Variables:

mllt_iterations (list) – List of iterations to perform MLLT estimation

calc_lda_mllt()

Multiprocessing function that calculates LDA+MLLT transformations.

See also

CalcLdaMlltFunction

Multiprocessing helper function for each job

LdaTrainer.calc_lda_mllt_arguments

Job method for generating arguments for the helper function

bin/est-mllt.cc

Relevant Kaldi binary

gmmbin/gmm-transform-means.cc

Relevant Kaldi binary

featbin/compose-transforms.cc

Relevant Kaldi binary

train_lda_mllt.sh

Reference Kaldi script

calc_lda_mllt_arguments()

Generate Job arguments for CalcLdaMlltFunction

Returns:

Arguments for processing

Return type:

list[CalcLdaMlltArguments]

compute_calculated_properties()

Generate realignment iterations, MLLT estimation iterations, and initial gaussians based on configuration

lda_acc_stats()

Multiprocessing function that accumulates LDA statistics.

See also

LdaAccStatsFunction

Multiprocessing helper function for each job

LdaTrainer.lda_acc_stats_arguments

Job method for generating arguments for the helper function

bin/est-lda.cc

Relevant Kaldi binary

train_lda_mllt.sh

Reference Kaldi script

lda_acc_stats_arguments()

Generate Job arguments for LdaAccStatsFunction

Returns:

Arguments for processing

Return type:

list[LdaAccStatsArguments]

property lda_options

Options for computing LDA

train_iteration()

Run a single LDA training iteration

property train_type

Training identifier