Click here to register.

Speech Recognition Engines

Flat
sphinx_fe
User: raj
Date: 6/18/2009 10:00 pm
Views: 5631
Rating: 11

I following the Adapting the default acoustic model documentation. 
 
so I record 20 wav files with my voice and I also create transcription file, listoffiles. 
 
an4.txt and an4.dic taken from an4 database. 
 
my question is how to generate mfcs feature files. 
 
I think it is same as preliminary training by using SphinxTrain and perl scripts. 
 
OR some other else plz tell me. 
 
And sphinx_fe tool also not understand. 
 
sphinx_fe tool is generate a set of acoustic model feature file for raw audio recordings ,but where this tool run.clear directory stucture is not their. 
 
 
I follows directory structure as follows 
 
adapt/an4.txt 
/an4.dic 
/an4_train.fileids 
/an4_train.transcription 
/an4_01.wav 



/an4_20.wav 
/bw 
/mllr_solve 
 
bw and mllr_solve files taken from SphinxTrain/bin/release. 

--- (Edited on 6/18/2009 10:00 pm [GMT-0500] by Visitor) ---

Re: sphinx_fe
User: kmaclean
Date: 6/22/2009 11:31 am
Views: 76
Rating: 11

Hi raf,

>my question is how to generate mfcs feature files.

from the CMU Robust Group Tutorial (Learning to use the CMU SPHINX Automatic Speech Recognition system):

To compute MFCCs from the wave files, from the top level directory, namely an4, type the following from the command line:

             perl scripts_pl/make_feats.pl -ctl etc/an4_test.fileids

Ken

 

--- (Edited on 6/22/2009 12:31 pm [GMT-0400] by kmaclean) ---

Re: sphinx_fe
User: Raj
Date: 6/23/2009 8:10 am
Views: 226
Rating: 10

 

   Hi  kmaclean

    Thanks for given reply

     I followed sphinx robust group tutorial and done the following

    - generating mfcc for training data

    -Training an4 database

    -generate mfcc for testing data

    -testing an4 database ie decoding is also over

  I get word error get rate.

Now I want to do speaker adaptation. I following below link doc

   http://www.speech.cs.cmu.edu/cmusphinx/moinmoin/AcousticModelAdaptation

 here I am not understand directory structure

  I record 20 wav files with my voice and I also create transcription file, listoffiles.
  
  an4.txt and an4.dic taken from an4 database.

  my question is how to generate mfcs feature files.

  I think it is same as preliminary training by using SphinxTrain and perl scripts.

  OR some other else plz tell me.

 And sphinx_fe tool also not understand.
 
    sphinx_fe tool is generate a set of acoustic model feature file for raw audio recordings ,but where this tool run.clear directory stucture is not their.

 
       I follows directory structure as follows

         adapt/an4.txt
              /an4.dic
              /an4_train.fileids
              /an4_train.transcription
              /an4_01.wav
              .
              .
              .
              /an4_20.wav
              /bw
              /mllr_solve

     bw and mllr_solve files taken from SphinxTrain/bin/release.

Atleast u have tell difference between Training and adaptation.

--- (Edited on 6/23/2009 8:10 am [GMT-0500] by Visitor) ---

Re: sphinx_fe
User: kmaclean
Date: 7/2/2009 3:28 pm
Views: 1777
Rating: 10

Hi Raj,

Sorry, I am not familiar with speaker adaptation in Sphinx,

Ken

--- (Edited on 7/2/2009 4:28 pm [GMT-0400] by kmaclean) ---

PreviousNext