Menu

Problem recognizing female voice

Help
Madhukar
2019-04-10
2019-04-14
  • Madhukar

    Madhukar - 2019-04-10

    Hi,

    I'm using Pocketsphinx for keyword spotting. I followed the procedure
    mentioned " Adapting a default acoustic model" (
    https://cmusphinx.github.io/wiki/tutorialadapt/) to improve the accuracy by
    building a model with as many audio sample as possible (which includes both
    male and female voices). Male voices are recognized properly but it does
    not recognize any of the female voices. Do I build the model with more
    female voice samples? Is there any other alternative to recognize female
    voices?

    Thanks,
    Madhukar

     
    • Nickolay V. Shmyrev

      Adaptation does not help with keyword spotting, you have to train model from scratch probably and better focus on some more advanced algorithm (depends on your goals).

       
  • Madhukar

    Madhukar - 2019-04-12

    Thanks for the reply.

    Will this URL mentioned below help me.
    https://cmusphinx.github.io/wiki/tutorialam/

    What is the minimum data (number of hours of recording) required to go with training the model.

     
  • Madhukar

    Madhukar - 2019-04-12

    I'm trying to use Pocketsphinx only for detection for one keyword. Something like "hello champ". Detection with all genders. Do I still need to train model from scratch and have the following data requirements met (from documentation).

    AND you have plenty of data to train on:
    1 hour of recording for command and control for a single speaker
    5 hours of recordings of 200 speakers for command and control for many speakers
    10 hours of recordings for single speaker dictation
    50 hours of recordings of 200 speakers for many speakers dictation

     
    • Nickolay V. Shmyrev

      Do I still need to train model from scratch and have the following data requirements met (from documentation).

      Yes.

      I'm trying to use Pocketsphinx only for detection for one keyword. Something like "hello champ".

      You can also try https://github.com/MycroftAI/mycroft-precise, but you still have to train it.

       
  • Madhukar

    Madhukar - 2019-04-14

    Thanks for the reply.

    Currently I do not have enough data. To start with is it enough to have 1 hr of recording of two different speakers (Both command and dictation).

     
    • Nickolay V. Shmyrev

      You can start of course. Practical implementation requires much more data.

       

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.