CODEWITHSUNDEEP
speech-recognitionspeech-to-textspeech
Pravin Ramteke
Pravin Ramteke

Reputation: 63

How to use multiple features such as pitch, formants, shimmer and jitter to train HMM using HTK toolkit instead of MFCC only?

I am trying to develop a note transcription system using HMM. For this I am using HTK Toolkit. I want to use different features such as pitch, formants, shimmer and jitter instead of MFCC only.

How can I create a feature vector of above features to train HMM in HTK so that I can improve the performance of the system?

Upvotes: 0

Views: 587

Answers (1)

Nikolay Shmyrev
Nikolay Shmyrev

Reputation: 25220

This issue has been covered many times on HTK forums and mailing list, you can search for more details.

Basically you need to create your features first for every frame in some form, for example in text table. Then you can convert those values to a binary HTK format using C code or Matlab code. C code is not very complex but it requires you an understanding of HTK binary format. For example you can check detailed code at:

http://blog.jamesrossiter.co.uk/2008/11/16/converting-csv-and-vector-data-to-native-htk-format-using-c/

http://www.ee.ic.ac.uk/hp/staff/dmb/voicebox/doc/voicebox/writehtk.html

Upvotes: 1

Related Questions