Acoustic model adaptation problems

[Akshay]

I have been trying to do the adaptation process on acoustic model, but I get stuck at the 'bw' part everytime.

I am able to extract .mfc from .wav using sphinx_fe and convert mdef to text. I don't think I am doing anything wrong with these commands, but please correct me if I am:

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

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

This is how I run the bw program:

bw -hmmdir en-us -moddeffn en-us/mdef.txt -ts2cbfn .ptm. -feat ls_c_d_dd -svspec 0-12/13-25/26-38 -cmn current -agc none -dictfn cmudict-en-us.dict -ctlfn adapt.fileids -lsnfn adapt.transcription -accumdir .

OUTPUT:

INFO: main.c(229): Compiled on Jan 24 2016 at 08:16:10
Current configuration:
[NAME] [DEFLT] [VALUE]
-2passvar no no
-abeam 1e-100 1.000000e-100
-accumdir .
-agc none none
-agcthresh 2.0 2.000000e+000
-bbeam 1e-100 1.000000e-100
-cb2mllrfn .1cls. .1cls.
-cepdir
-cepext mfc mfc
-ceplen 13 13
-ckptintv 0
-cmn current current
-cmninit 8.0 8.0
-ctlfn adapt.fileids
-diagfull no no
-dictfn cmudict-en-us.dict
-example no no
-fdictfn
-feat 1s_c_d_dd ls_c_d_dd
-fullsuffixmatch no no
-fullvar no no
-help no no
-hmmdir en-us
-latdir
-latext
-lda
-ldadim 0 0
-lsnfn adapt.transcription
-lw 11.5 1.150000e+001
-maxuttlen 0 0
-meanfn
-meanreest yes yes
-mixwfn
-mixwreest yes yes
-mllrmat
-mmie no no
-mmie_type rand rand
-moddeffn en-us/mdef.txt
-mwfloor 0.00001 1.000000e-005
-npart 0
-nskip 0
-outphsegdir
-outputfullpath no no
-part 0
-pdumpdir
-phsegdir
-phsegext phseg phseg
-runlen -1 -1
-sentdir
-sentext sent sent
-spthresh 0.0 0.000000e+000
-svspec 0-12/13-25/26-38
-timing yes yes
-tmatfn
-tmatreest yes yes
-topn 4 4
-tpfloor 0.0001 1.000000e-004
-ts2cbfn .ptm.
-varfloor 0.00001 1.000000e-005
-varfn
-varnorm no no
-varreest yes yes
-viterbi no no

INFO: feat.c(715): Initializing feature stream to type: 'ls_c_d_dd', ceplen=13,
CMN='current', VARNORM='no', AGC='none'
FATAL: "feat.c", line 845: Bad feature type argument

I don't know how to resolve the 'bad feature type argument' error.

Also, I would like to know how this bw program contributes to the adaptation process. Since, it doesn't produce any output file, I assumed I could skip it and proceed with map_adapt/mllr_solve part. But that gives me some error too. Is it because it is mandatory to run bw program before map_adapt/mllr_solve? Will I be able to run them properly if I am able to run bw properly?

You can have a look at the acoustic model I am using and my adaptation corpus at: https://docs.google.com/uc?id=0B221TxAHKuBSdFlveWZOaXRsU0k&export=download

[Nickolay V. Shmyrev]

INFO: feat.c(715): Initializing feature stream to type: 'ls_c_d_dd', ceplen=13,
CMN='current', VARNORM='no', AGC='none'
FATAL: "feat.c", line 845: Bad feature type argument

Feature type must be 1s_c_d_dd with 1 (one) instead of l (L letter).

[Akshay]

Thank you so much. It works now.
I used map_adapt for adaptation.

Does the language model and the dictionary get adapted accordingly? If not, how can I adapt language model or add words to dictionary?

[Nickolay V. Shmyrev]

If not, how can I adapt language

http://cmusphinx.sourceforge.net/wiki/tutoriallmadvanced

or add words to dictionary?

http://cmusphinx.sourceforge.net/wiki/tutorialdict