The error rate is even higher with adapted acoustic modeling that default en-us model!

[Freya]

The error rate is even higher with adapted acoustic modeling than default en-us model!

Hi, I created an adapated acoustic modeling, following the instructions http://cmusphinx.sourceforge.net/wiki/tutorialadapt. I included timit voices in the adapted model.

However, the word error rate was 65.71%. If I just used the default database of sphinx, the word error rate was about 50%. Does that make sense? Why does the word error rate become higher if I train it? Thanks in advance.

[Nickolay V. Shmyrev]

You did not provide enough information to get help on this issue.

[Freya]

I have attached the files: en-us-adapt, test.hyp, test and train files. I used the an4 examples for training and testing.

Thanks.

[Nickolay V. Shmyrev]

The data you provided is not complete. I do not see the decoding command, decoding logs, I also do not see any "timit voices" which you described above.

[Freya]

Sorry. I used the an4 examples. But it is still very low.

[Freya]

Here are all the commands I used.

sphinx_fe -argfile en-us/feat.params -samprate 16000 -c train.fileids -di . -do . -ei wav -eo mfc -mswav yes

pocketsphinx_mdef_convert -text en-us/mdef en-us/mdef.txt

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


./mllr_solve -meanfn en-us/means -varfn en-us/variances -outmllrfn mllr_matrix -accumdir .


cp -a en-us en-us-adapt


./map_adapt \
    -moddeffn en-us/mdef.txt \
    -ts2cbfn .semi. \
    -meanfn en-us/means \
    -varfn en-us/variances \
    -mixwfn en-us/mixture_weights \
    -tmatfn en-us/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



./mk_s2sendump \
    -pocketsphinx yes \
    -moddeffn en-us-adapt/mdef.txt \
    -mixwfn en-us-adapt/mixture_weights \
    -sendumpfn en-us-adapt/sendump


./pocketsphinx_continuous -hmm en-us-adapt -lm en-us.lm.dmp -dict cmudict-en-us.dict



pocketsphinx_batch \
 -adcin yes \
 -cepdir wav \
 -cepext .wav \
 -ctl test.fileids \
 -lm en-us.lm.dmp \
 -dict cmudict-en-us.dict \
 -hmm en-us-adapt \
 -hyp test.hyp


./word_align.pl test.transcription test.hyp

[Nickolay V. Shmyrev]

The model you used is ptm model, not semi model. I have no idea why you replaced ptm with semi.