Menu

Recognition accuracy using MLLR

Help
Balaji
2016-11-01
2016-11-08
  • Balaji

    Balaji - 2016-11-01

    Hello Nickolay,

    I am using "Adapting existing acoustic model" section of the cmusphinx tutorial to improve recognition accuracy. I have communicated earlier using the forum post "How to improve the recognition accuracy using mllr". Presently, I am using only one voice (female, adult, Indian English) to train and test the system. Each time I run the transcriber.java, I observe the recognition accuracy is improving, even through I have not increased the vocabulary (the amount of training data).
    Is there a way that the system is learning to improve the recognition accuracy? I have run the steps in the tutorial and created an mllr_matrix. Using the same, I have been able to achieve even 100% accuracy for some of the sentences.

    I am concerned that there is something I am missing in the testing part. My ultimate aim is to recognize free form speech. Are there standard test cases available to test the recognition accuracy? Kindly let me know.

    Thank you.

    Balaji.

     
    • Nickolay V. Shmyrev

      Each time I run the transcriber.java, I observe the recognition accuracy is improving, even through I have not increased the vocabulary (the amount of training data).

      It is not quite clear what is your issue here.

      I am concerned that there is something I am missing in the testing part.

      Talk to a psychologist about that.

       

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.