calc_fmllr

aligner.multiprocessing.calc_fmllr(directory, split_directory, sil_phones, num_jobs, config, initial=False, iteration=None)[source]

Multiprocessing function that computes speaker adaptation (fMLLR)

See:

for more details on the Kaldi binary this runs.

Also see https://github.com/kaldi-asr/kaldi/blob/master/egs/wsj/s5/steps/align_fmllr.sh for the original bash script that this function was based on.

Parameters:
directorystr

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

split_directorystr

Directory of training data split into the number of jobs

sil_phonesstr

Colon-separated list of silence phones

num_jobsint

The number of processes to use in calculation

configTriphoneFmllrConfig

Configuration object for training

initialbool, optional

Whether this is the first computation of speaker-adaptation, defaults to False

iterationint

Specifies the current iteration, defaults to None