TriphoneTrainer

class montreal_forced_aligner.acoustic_modeling.TriphoneTrainer(subset=5000, num_iterations=35, num_leaves=1000, max_gaussians=10000, cluster_threshold=-1, boost_silence=1.25, power=0.25, **kwargs)

Bases: AcousticModelTrainingMixin

Triphone trainer

Parameters:

• subset (int) – Number of utterances to use, defaults to 5000

• num_iterations (int) – Number of training iterations to perform, defaults to 35

• num_leaves (int) – Number of states in the decision tree, defaults to 1000

• max_gaussians (int) – Number of gaussians in the decision tree, defaults to 10000

• cluster_threshold (int) – For build-tree control final bottom-up clustering of leaves, defaults to 100

See also

AcousticModelTrainingMixin

For acoustic model training parsing parameters

acoustic_model_training_params()

Configuration parameters

compute_calculated_properties()

Generate realignment iterations and initial gaussians based on configuration

convert_alignments()

Multiprocessing function that converts alignments from previous training

See also

ConvertAlignmentsFunction

Multiprocessing helper function for each job

TriphoneTrainer.convert_alignments_arguments

Job method for generating arguments for the helper function

train_deltas.sh

Reference Kaldi script

train_lda_mllt.sh

Reference Kaldi script

train_sat.sh

Reference Kaldi script

convert_alignments_arguments()

Generate Job arguments for ConvertAlignmentsFunction

Returns:

Arguments for processing

Return type:

list[ConvertAlignmentsArguments]

property phone_type

Phone type

property train_type

Training identifier

tree_stats()

Multiprocessing function that computes stats for decision tree training.

See also

tree_stats_func

Multiprocessing helper function for each job

TriphoneTrainer.tree_stats_arguments

Job method for generating arguments for the helper function

bin/sum-tree-stats.cc

Relevant Kaldi binary

train_deltas.sh

Reference Kaldi script

train_lda_mllt.sh

Reference Kaldi script

train_sat.sh

Reference Kaldi script

tree_stats_arguments()

Generate Job arguments for tree_stats_func

Returns:

Arguments for processing

Return type:

list[TreeStatsArguments]