Menu

sphinx train

2002-04-07
2023-09-13
  • Stefano Canepa

    Stefano Canepa - 2002-04-07

    Dear all,
       I am trying to add a little Italian to sphinx. I am using sphinxtrain. I'm following the manual. But my problem is that the .wav file I recorded using the GNOME sound recorder cannot be converted in feature files.
       Could you help me. I'm thinking it's an hardware problem.

       Stefano
       sc@linux.it
      

     
    • Anonymous

      Anonymous - 2002-04-08

      Whats the reported problem ? Can you play the sound files back..are they okay...

       
    • Doumenc Julie

      Doumenc Julie - 2002-04-15

      Hi,

      I'm trying to create my own LM in english (I'm French so I try to add the French accent to the model).
      I followed the instructions of doc/tinydoc.txt of SphinxTrain. My problem is that the command: bin/make_feats etc/time.fileids doesn't give any result, even if it uses 95% of the CPU. I waited for 48 hours... First, I tried with 132 sentences (20 minutes of recording) and as it was very long, I stopped it and tried with 10 seconds of recording but the program never stopped...

      Do you have an idea of the problem?

      Thanks a lot

      julie

       
    • Doumenc Julie

      Doumenc Julie - 2002-04-17

      OK, I found the solution: I changed the options of the command line: "bin/wave2feat -verbose -c $1 -nist -di wav -ei wav -do feat -eo feat" to "bin/wave2feat -verbose -c $1 -raw -di wav -ei wav -do feat -eo feat"
      And my .feat files were generated!

      Now, I have an other problem: the command "./script_pl/02.ci_schmm/slave_convg.pl" gives :
      WARN: "s3io.c", line 253: Unable to open /home/doumenc/SphinxTrain/time/model_parameters/time.ci_semi_flatinitial/mixture_weights for reading; No such file or directory.

      Indeed, this file doesn't exist. But I thought that slave_convg.pl had to create it and write in it.

      Any ideas?

       
      • prathna

        prathna - 2023-09-13

        hello i am having same error..can you help me
        Can not open transcript file (/home/prathna/speech_recognition/other/etc/kreol_train.transcription) at /home/prathna/speech_recognition/sphinxtrain/scripts/00.verify/verify_all.pl line 245.

         
    • Anonymous

      Anonymous - 2002-04-18

      Hello again,

      Go look at the log files carefully. Any failure at a stage is usually caused by a failure at a previous stage. Use a html browser to view time.html file.

       
    • Stefano Canepa

      Stefano Canepa - 2002-04-27

      Doumenc Julie and Andrew Hewitt I need to thank you pubblically for you help. I came really late becouse I was monitoring the wrong forum and becouse I was abroad and away from my PC a while.

      Thanks
      Stefano

       
    • datta prasad

      datta prasad - 2002-08-22

      hi all
      this is dattu. i m a fresher to this Sphinx software.i was able to get feat files but Phase 01 is giving trouble.
      actually i am runing the scripts described in the tinydoc.txt.
      ./scripts_pl/01.vector_quantize/slave.VQ.pl

      running the above script , the program is not returning. i left that all the night but still its like that.
      i saw the log files
      i got an error saying
      1) NO mdef files
      2) line no1271 unable to open dumpfile  ....../time.dmp for opening
      3) line no 1549 unabel to train
      could any one help me in this issue Pls.

       
      • Jessica P. Hekman

        dattu,

        Hopefully someone who actually knows what they are talking about will try to help you, but in the meantime I am willing to take a shot at it if you can post your actual log file instead of just a summary.

        As for the error that time.dmp can't be written -- have you checked to make sure there are no permissions problems?

         
    • datta prasad

      datta prasad - 2002-08-23

      thanks jessica,
      i had seen permissions , its ok
      any how i getting the following problem

      at phase 01.vector quantize , there r two steps to be done 1) AGG_SEG and kmeans

      error is occurring at AGG_SEG. the exact one is as follows
      INFO: main.c (168): No lexical transcripts provided
      INFO: corpus.c(426): Will process all remaining utts starting at 0
      INFO: main.c (272): Will produce feat files
      INFO: main.c(426):Writing frames to one file
      Header size field:  -411172864(e77e0000): filesize: 129953(1fba1)
      ERROR: corpus.c line (1513): MFCC read failed. Retrying after sleep...

      could u help me Pls

       
    • Anonymous

      Anonymous - 2002-08-30

      g_dattu:

      I have had this problem myself.  Usually it means that the system can't find the MFC file associated with (at least) one of the files in the control file.
        -make sure you have run wave2feat to convert your audio to mel-cepstral feature files
        -Check the value of CFG_FEATFILE_EXTENSION and make sure it matches the extension on your mfc files (from wave2feat)
        -make sure $CFG_FEATFILES_DIR actually points to the location where the mfc files are

      The reason it never ends is because when an mfc read fails, there is a sleep command in one of the libraries (libio.lib maybe).  I actually changed that to an exit and added a couple of lines to print the file that caused the problem.  I can't, unfortunately, give the actual code (it's my company's), but what you can do is write a perl script or something that verifies an mfc file exists for every file listed in your control file.

      --Keith

       
    • Anonymous

      Anonymous - 2002-09-17

      I am training a word model, but the script agg_seg.pl goes to sleep and logs the following output:

      INFO: main.c(168): No lexical transcripts provided
      INFO: corpus.c(1236): Will process all remaining utts starting at 0
      INFO: main.c(272): Will produce FEAT dump
      INFO: main.c(426): Writing frames to one file
      stat_retry(/sphinxtrain/bg_tel/feat/blago_0_1
      .feat) failed
      ERROR: "corpus.c", lin

      does anyone know a reason for this?

      this feat file is 3072 bytes -> could this be the reason(too small)
      thanks

       
    • Anonymous

      Anonymous - 2002-09-24

      Hi plamen (first name?  Russian or some other slavic)?.

      Generally, this is a result of the mfc file not existing (see just previous post).  If not (I am sure it isn't :) ), did you use wave2feat to produce an mfc/feat file?  If you ported wave2feat to windows, there is at least one case where it loads the original audio file as a text file when it should be a binary.  Do you have permission to write to the directory involved (is it specified as read-only) and the file doesn't already exist?

       
    • dasineni

      dasineni - 2008-04-19

      hi Doumenc Julie
      you replied like this

      OK, I found the solution: I changed the options of the command line: "bin/wave2feat -verbose -c $1 -nist -di wav -ei wav -do feat -eo feat" to "bin/wave2feat -verbose -c $1 -raw -di wav -ei wav -do feat -eo feat"
      And my .feat files were generated!

      can u please explain how to run this command after creating the dic file(by following doc/tinydoc.txt of SphinxTrain)for generating the feat files

      thanks for your help

      dasineni..

       

Log in to post a comment.

Want the latest updates on software, tech news, and AI?
Get latest updates about software, tech news, and AI from SourceForge directly in your inbox once a month.