This is the command I used for training en-us-orig(earlier it was en-us) but I am getting the errors like this..
"Read en-us-orig/mixture_weights [5126x3x128 array]
FATAL: "mod_inv.c", line 358: Number of feature streams in mixture_weights file 3 differs from the configured value 1, check the command line options
"
I tried changing the etc/sphinx_config and increased $CFG_FEAT_PARAMS but I am getting errors like this
Read en-us-orig/mixture_weights [5126x3x128 array]
FATAL: "mod_inv.c", line 358: Number of feature streams in mixture_weights file 3 differs from the configured value 1, check the command line options.
PLEASE HELP
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
The model you are trying to adapt is ptm model, not continuous model. The adaptation of this model is properly covered by adaptation tutorial, you need to follow it step-by-step without any modifications.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
I had used the starting model from sphinx4-5prealpha src which contained the continuous model. But it(en-us) inside the src folder doesn't contain the feature_trasform folder..which is usually present in continuous models.
So does this src contain the continuous model folder. If not, can you send me a download link which works?
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
WARN: "accum.c", line 628: Over 500 senones never occur in the input data. This is normal for context-dependent untied senone training or for adaptation, but could indicate a serious problem otherwise.
Will this impact the accuracy??
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
./bw \
-hmmdir en-us-orig \
-moddeffn en-us-orig/mdef.txt \
-ts2cbfn .cont. \
-feat 1s_c_d_dd \
-cmn current \
-agc max \
-dictfn cmudict-en-us.dict \
-ctlfn wav_recordings.fileids \
-lsnfn wav_recordings.transcription \
-accumdir .
This is the command I used for training en-us-orig(earlier it was en-us) but I am getting the errors like this..
"Read en-us-orig/mixture_weights [5126x3x128 array]
FATAL: "mod_inv.c", line 358: Number of feature streams in mixture_weights file 3 differs from the configured value 1, check the command line options
"
I tried changing the etc/sphinx_config and increased $CFG_FEAT_PARAMS but I am getting errors like this
Read en-us-orig/mixture_weights [5126x3x128 array]
FATAL: "mod_inv.c", line 358: Number of feature streams in mixture_weights file 3 differs from the configured value 1, check the command line options.
PLEASE HELP
The model you are trying to adapt is ptm model, not continuous model. The adaptation of this model is properly covered by adaptation tutorial, you need to follow it step-by-step without any modifications.
Hi,
I had used the starting model from sphinx4-5prealpha src which contained the continuous model. But it(en-us) inside the src folder doesn't contain the feature_trasform folder..which is usually present in continuous models.
So does this src contain the continuous model folder. If not, can you send me a download link which works?
You can download continuous model in files:
http://sourceforge.net/projects/cmusphinx/files/Acoustic%20and%20Language%20Models/US%20English%20Generic%20Acoustic%20Model/cmusphinx-en-us-5.2.tar.gz/download
Thanks Nickolay, it worked.
BTW..
WARN: "accum.c", line 628: Over 500 senones never occur in the input data. This is normal for context-dependent untied senone training or for adaptation, but could indicate a serious problem otherwise.
Will this impact the accuracy??
*feature_transform file