Menu

sphinx2-test difficulties

Help
Anonymous
2002-11-30
2012-09-22
  • Anonymous

    Anonymous - 2002-11-30

    I am new to the Sphinx program, and I am trying to use the sphinx2-test script to decode a file other than goforward.16k. I recorded a couple of wav files in the following format:

    16bit PCM
    16000Hz sampling rate
    mono

    I saved the files with the extension .16k. I then changed the turtle.ctl file to specify the name of the file I wanted to use. When I run the script, I get the following output:
    When audio file contains the utterance "You are home" it is decoded as "You are halt." I also tried to record the phrase "Go forward ten meters" in a file of my own and it decoded as "Go one right a meters."

    Do you have any ideas on what might be wrong? (If the answer is removing the 44 byte header from the wav file, please let me know how. Also, do I need to change any of the arguments used in sphinx2-test?)

    I would greatly appreciate any help you can give me!

     
    • robert b

      robert b - 2002-12-02

      I'd like to be clear what you are doing.  Are you training with a WAV file that contains "You are home" and then decoding using the same WAV file?

       
    • Anonymous

      Anonymous - 2002-12-03

      I am recording a new WAV file, and trying to use it instead of goforward.16k in the sphinx2-test shell script.

      I have not done any training - maybe that is the problem (I am very new to this program and speech recognition technology in general). Could you please let me know how I can do voice training with Sphinx?

      Thanks in advance.

       
    • Anonymous

      Anonymous - 2002-12-04

      I will download SphinxTrain, but I was under the impression that it is used to create new acoustic models. I am just trying to get Sphinx to recognize a phrase that already exists in the default language model included with the distribution ("Go forward ten meters" or "I am home" or "one, two, three, four, five"). The words I am using are in the dictionary of this model, shouldn't they be recognized without using SphinxTrain?

       
      • robert b

        robert b - 2002-12-04

        Well, you seemed to indicate that you wanted to replace goforward.16k with another WAV file.  Just replacing it won't do anything -- you need to do a full training.  SphinxTrain works by encoding WAV files into Sphinx3-format model files, and then those are converted to Sphinx2-format CHMMs which can then be used by sphinx2-test or whatever.  See http://xvoice.arborius.net/xvoice-sphinx/FrontPage for specific directions on adapting existing models.

         
    • Anonymous

      Anonymous - 2002-12-04

      Thank you very much for your time and help!

      I have just one more question: Does this mean that in order for sphinx2-test or sphinx2-continuous or any of the other programs to recognize a phrase I will need to do a full training through SphinxTrain, and if so, after the training, will it only work for my voice or can it recognize multiple voices?

      Thanks again.

       
      • robert b

        robert b - 2002-12-04

        In theory, if you _adapt_ an existing model to your voice, it will work better with your voice, but should work (but less well) with others voices.  I don't think anyone has actually tried this with the turtle stuff -- we've only tried (with limited success) with hub4.

        If you manage to successfully adapt turtle, please post a message to this forum -- I'd be curious.

        You can also try, instead of adapting an existing model, creating your own model from scratch.  In this case, the result will only work with your voice and no others.  You need to use SphinxTrain in the standard way -- see "doc/tinydoc.txt" in the SphinxTrain directory on how to do this.  Various people have had mixed success with this, probably due to short transcriptions -- longer are usually better.

         
    • Anonymous

      Anonymous - 2002-12-04

      Thank you very much for all your time and help. I will let you know if I make any progress with the turtle model.

       

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.