Menu

Multicondition Training with CMU Sphinx

Help
2018-08-21
2018-08-23
  • Ramanathan Rahul

    Hi
    I am working on CMU Sphinx for speech to text conversion. I have 100 hours of clean speech with which I have integrated some noise, so that I have 100 hours of clean + 100 hours of noisy speech. I want to perform multicondition training (i.e., fetch alignments of only clean speech & map with noisy speech). Want to know if such a thing is already in-built in CMU sphinx or not?

     
    • Nickolay V. Shmyrev

      Want to know if such a thing is already in-built in CMU sphinx or not?

      Such thing is not available. Also GMM algorithms of cmusphinx do not play well with multiconditional training as neural networks.

       
  • Ramanathan Rahul

    There are a couple of questions I have:
    1. How do I apply multi conditional training in CMU Sphinx?
    2. Secondly, I have trained CMU with only clean speech & I am trying to decode with pocketsphinx. It shows 3 errors: bigram are not in unigram; fail to read lm file; an4.ug.lm.dmp (my lm file) is not a dump file. Need help in sorting this out.

     
    • Nickolay V. Shmyrev

      How do I apply multi conditional training in CMU Sphinx?

      Create a multiconditional dataset and train on it as usual.

      It shows 3 errors: bigram are not in unigram; fail to read lm file; an4.ug.lm.dmp (my lm file) is not a dump file. Need help in sorting this out.

      This forum has search.

       
      • Ramanathan Rahul

        How do I create the multi conditional dataset? Do I save both clean and noisy speech in the same folder? Like if I have a folder 1 with a file id 1-1111-0000.wav, do I store the noisy file of this recording say 1-1111-0000_noisy.wav in the same folder 1?

         
        • Ramanathan Rahul

          I have these 2 doubts:
          1. How do I create the multi conditional dataset? Do I save both clean and noisy speech in the same folder? Like if I have a folder 1 with a file id 1-1111-0000.wav, do I store the noisy file of this recording say 1-1111-0000_noisy.wav in the same folder 1?
          2. Do I need some separate code also for multi condition training or just a multi condition dataset will do? And those errors which I've mentioned earlier, I haven't found a suitable solution yet in the forum. So can you help with a solution or link?

           
  • Ramanathan Rahul

    Do I need some separate code also for multi condition training or just a multi condition dataset will do? And those errors which I've mentioned earlier, I haven't found a suitable solution yet. So can you help with a solution?

     

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.