I am working in text independet speaker identfication in a closed set
environment. I have used 19 MFCC feature extracted using 20 triangular
shaped linearly spaced filter in mel scale to model speakers using
GMM. I want to make a study on different shapes of
filter(triangular .
Can you suggest me a good starting paper that contains a review talk
about filtershape, number of filters in filterbank, spacing of the
filter, number of co-efficients required etc.
Thanking You
Md. Sahidullah
I think the difference is really unsufficient. Google gives enough:
http://slt.wcl.ee.upatras.gr/papers/ganchev17.pdf
http://maxwell.me.gu.edu.au/spl/publications/papers/merc03_ben.pdf
http://www.cnel.ufl.edu/~markskow/papers/iscas03.pdf
For speaker identification there is a freely available toolkit -
ALIZE. It's interesting that authors suggest to use different features
extracted with SPro4 than HTK's mfcc to get better recognition rate:
http://mistral.univ-avignon.fr/pdf/article_989-alize_odyssey08.pdf
Thank You Very much for your response. I have read the papers one by
ganchev and another by paliwal sir. But, they did not give me
information which I had exactly wanted. Thank you very much for
providing information about ALIZE. I am checking it.
Regards
Md. Sahidullah