TrainableAligner

class montreal_forced_aligner.acoustic_modeling.trainer.TrainableAligner(training_configuration=None, phone_set_type=None, model_version=None, **kwargs)

Bases: TranscriberMixin, TopLevelMfaWorker, ModelExporterMixin

Train acoustic model

Parameters:

• training_configuration (list[tuple[str, dict[str, Any]]]) – Training identifiers and parameters for training blocks

• phone_set_type (str) – Type of phone set to use for acoustic modeling

See also

CorpusAligner

For dictionary and corpus parsing parameters and alignment parameters

TopLevelMfaWorker

For top-level parameters

ModelExporterMixin

For model export parameters

Variables:

• param_dict (dict[str, Any]) – Parameters to pass to training blocks

• final_identifier (str) – Identifier of the final training block

• current_subset (int) – Current training block’s subset

• current_acoustic_model (AcousticModel) – Acoustic model to use in aligning, based on previous training block

• training_configs (dict[str, AcousticModelTrainingMixin]) – Training blocks

add_config(train_type, params)

Add a trainer to the pipeline

Parameters:

• train_type (str) – Type of trainer to add, one of monophone, triphone, lda or sat

• params (dict[str, Any]) – Parameters to initialize trainer

Raises:

ConfigError – If an invalid train_type is specified

align()

Multiprocessing function that aligns based on the current model.

See also

AlignFunction

Multiprocessing helper function for each job

AlignMixin.align_arguments

Job method for generating arguments for the helper function

align_si.sh

Reference Kaldi script

align_fmllr.sh

Reference Kaldi script

property align_options

Alignment options

property alignment_model_path

Current alignment model path

compute_phone_pdf_counts()

Calculate the counts of pdfs corresponding to phones

property configuration

Configuration for the worker

property data_directory

Current data directory based on the trainer’s subset

classmethod default_training_configurations()

Default MFA training configuration

export_files(output_directory, output_format=None, include_original_text=False)

Export a TextGrid file for every sound file in the dataset

Parameters:

• output_directory (Path) – Directory to save to

• output_format (str, optional) – Format to save alignments, one of ‘long_textgrids’ (the default), ‘short_textgrids’, or ‘json’, passed to praatio

• include_original_text (bool) – Flag for including the original text of the corpus files as a tier

export_model(output_model_path)

Export an acoustic model to the specified path

Parameters:

output_model_path (str) – Path to save acoustic model

property meta

Metadata about the final round of training

property model_path

Current model path

property num_current_utterances

Number of utterances in the current subset

classmethod parse_parameters(config_path=None, args=None, unknown_args=None)

Parse configuration parameters from a config file and command line arguments

Parameters:

• config_path (Path, optional) – Path to yaml configuration file

• args (dict[str, Any]) – Parsed arguments

• unknown_args (list[str]) – Optional list of arguments that were not parsed

Returns:

Dictionary of specified configuration parameters

Return type:

dict[str, Any]

setup()

Setup for acoustic model training

train()

Run through the training configurations to produce a final acoustic model

transition_acc_arguments()

Generate Job arguments for TransitionAccArguments

Returns:

Arguments for processing

Return type:

list[TransitionAccArguments]

property working_directory

Working directory

property working_log_directory

Current log directory