adapting-acoustics-models-steps

1. Generating acoustic feature files

sphinx_fe -argfile ../TrainingSet/acoustic-model/feat.params \
-samprate 16000 -c ted.fileids \
-di . -do . -ei wav -eo mfc -mswav yes

audio file format must be mono 16k
can use the following bash commend to change the format

for filename in *.wav; do 
ffmpeg -i "$filename" -ac 1 -ar 16000 ./Test1/"$filename"; done

Reference Screenshot:
alt text

2. Accumulating observation counts

../TrainingSet/bw \
 -hmmdir ../TrainingSet/acoustic-model\
 -moddeffn ../TrainingSet/acoustic-model/mdef.txt\
 -ts2cbfn .ptm. \
 -feat 1s_c_d_dd \
 -svspec 0-12/13-25/26-38 \
 -cmn current \
 -agc none \
 -dictfn ../TrainingSet/cmudict-en-us.dict\
 -ctlfn ted.fileids \
 -lsnfn tag-ted.transcription \
 -accumdir .

Reference Screenshot:
alt text

If a word is not in the dictionary model, it will not be able to produce phonetic utterance; therefore, skip the utterance

alt text

3. Updating the acoustic model files with MAP

cp -a ../TrainingSet/acoustic-model en-us-adapt 
<br>
../TrainingSet/map_adapt \
    -moddeffn ../TrainingSet/acoustic-model/mdef.txt \
    -ts2cbfn .ptm. \
    -meanfn ../TrainingSet/acoustic-model/means \
    -varfn ../TrainingSet/acoustic-model/variances \
    -mixwfn ../TrainingSet/acoustic-model/mixture_weights \
    -tmatfn ../TrainingSet/acoustic-model/transition_matrices \
    -accumdir . \
    -mapmeanfn en-us-adapt/means \
    -mapvarfn en-us-adapt/variances \
    -mapmixwfn en-us-adapt/mixture_weights \
    -maptmatfn en-us-adapt/transition_matrices

Reference Screenshot:
alt text

reference: https://cmusphinx.github.io/wiki/tutorialadapt/