My task is create voice recognition of Russian words list. that's why I need to use Russian acoustic model. For best recognition result, I working on acoustic model adapting.
I follow documentation link , all seems good but when i run script :
./bw \
-hmmdir model \
-moddeffn model/mdef.txt
-ts2cbfn .ptm. \
-feat 1s_c_d_dd \
-svspec 0-12/13-25/26-38 \
-cmn current \
-agc none \
-dictfn zero_ru.dic \
-ctlfn lesniki.fields \
-lsnfn lesniki.transcription\
-accumdir .
I get next Errors:
INFO: s3mixw_Io.c(117): Read model/mixture_weights [4159x1x63 array]
FATAL: "mod_In.c", line 358: Number of feature streams in
mixture_weights file 1 differs from the configured value 3, check the
command line options
There is my projects and some links
project on git
This error means that configuration parameters used for bw are mismatched with configuration parameter the model was trained with. From the file it seems you are using mixture weights from continuous model, but you are trying to adapt like ptm model. For continuous models you do not need -svspec option but you need -lda feature_transform option instead. And -ts2cbfn must be .cont., not .ptm..
If you tried to adapt continuous model, you need to change parameters accordingly. Otherwise, you took mixture_weights file from the wrong model.