acc_stats

aligner.multiprocessing.acc_stats(iteration, directory, split_directory, num_jobs, config)[source]

Multiprocessing function that computes stats for GMM training

See http://kaldi-asr.org/doc/gmm-acc-stats-ali_8cc.html for more details on the Kaldi binary this runs.

Also see https://github.com/kaldi-asr/kaldi/blob/master/egs/wsj/s5/steps/train_mono.sh for the bash script this function was extracted from

Parameters:
iteration : int

Iteration to calculate stats for

directory : str

Directory of training (monophone, triphone, speaker-adapted triphone training directories)

split_directory : str

Directory of training data split into the number of jobs

num_jobs : int

The number of processes to use in calculation

fmllr : bool, optional

Whether the current training session is using fMLLR (speaker-adaptation), defaults to False