Menu

Tuning LM and DICT to correspond to the correct words

Help
2016-04-28
2016-04-28
  • Ananthapadmanabhan

    I am trying to develop sphinx to identify different accent without training the hmm file.

    So, what I did is:-
    1. took a large file (which had the phrase 'the effort right from the start are evident') and built a language model out of it
    2. uttered the sentence (the effort right from the start are evident), used phoneme recogniser to identify the phonemes in it. (Which came out as "D EH D AY UH D AE T AH D AY D AH Z EH R AH N IH Z IH HH UH D")
    3. now made a .dict file where only these 7 words and only corresponding phoneme (for ex. the is D EH, effort is D AY UH D and so on)
    4. tried to decode with pocketsphinx_continuous using the above built dict file and lm file

    But still I am not getting the output correctly as ('the effort ...) . Could anyone explain to me why this has not gotten trained to identify and how can I improve its efficiency.

     
    • Nickolay V. Shmyrev

      It is pretty senseless activity.

       
      • Ananthapadmanabhan

        I just wanted to verify that the construction of the text from the phonemes is accurate given lm and dictionary. Is there any other way to do it? And if it is inaccurate, then how would I go about improving it.

         
        • Nickolay V. Shmyrev

          Write proper dictionary and adapt the acoustic model.

           
          • Ananthapadmanabhan

            Thanks man. I will try building a dictionary according to my pronounciation and then adapting the model for it.

             

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.