AdaptingAligner#
- class montreal_forced_aligner.alignment.AdaptingAligner(mapping_tau=20, **kwargs)[source]#
Bases:
PretrainedAligner
,AdapterMixin
Adapt an acoustic model to a new dataset
- Parameters:
mapping_tau (int) – Tau to use in mapping stats between new domain data and pretrained model
See also
PretrainedAligner
For dictionary, corpus, and alignment parameters
AdapterMixin
For adapting parameters
- Variables:
- acc_stats(alignment=False)[source]#
Accumulate stats for the mapped model
- Parameters:
alignment (bool) – Flag for whether to accumulate stats for the mapped alignment model
- property align_directory#
Align directory
- property alignment_model_path#
Current acoustic model path
- export_model(output_model_path)[source]#
Output an acoustic model to the specified path
- Parameters:
output_model_path (str) – Path to save adapted acoustic model
- map_acc_stats_arguments(alignment=False)[source]#
Generate Job arguments for
AccStatsFunction
- Returns:
Arguments for processing
- Return type:
list[
AccStatsArguments
]
- property meta#
Acoustic model metadata
- property model_path#
Current acoustic model path
- property next_model_path#
Mapped acoustic model path
- train_map()[source]#
Trains an adapted acoustic model through mapping model states and update those with enough data.
See also
AccStatsFunction
Multiprocessing helper function for each job
AdaptingAligner.map_acc_stats_arguments
Job method for generating arguments for the helper function
- gmmbin/gmm-sum-accs.cc
Relevant Kaldi binary
- gmmbin/gmm-ismooth-stats.cc
Relevant Kaldi binary
- gmmbin/gmm-est.cc
Relevant Kaldi binary
- train_map.sh
Reference Kaldi script
- property working_log_directory#
Current log directory