2

I am currently doing a speaker verification project using hidden markov models no accurate results on voice signals yet, though i have tested the system to various data samples (not involved with voice).

I extracted the MFCC of the voice signals using scikits talkbox. I assumed that no parameters must be changed and that the default ones are already fit for such project. I am suspecting that my problem is within the vector quantization of the mfcc vectors. I chose kmeans as my algorithm using scipy's kmeans clustering function. I was wondering if there is a prescribed number of clusters for this kind of work. I originally set mine to 32. Sample rate of my voice files are 8000 and 22050. Oh additionally, I recorded them and manually removed the silence using Audacity.

Any suggestions?

Bobby
  • 31
  • 5
  • Also consider using more advanced clustering algorithms. MFCC vectors may be too high dimensional for k-means to return meaningful results. In high-dimensional data, k-means results were found to often be just as good as a random nearest-neighbor quantization. – Has QUIT--Anony-Mousse Mar 14 '14 at 01:35
  • Hi there, thanks for the feedback. I am wondering if you could suggest a few clustering algorithms? I'm thinking of using the LBG algorithm. Currently searching for some now. Again, thank you very much. – Bobby Mar 14 '14 at 02:58
  • Most will need a similarity measure. Do you have a good similarity measure for MFCCs? – Has QUIT--Anony-Mousse Mar 14 '14 at 05:27
  • Im afraid I have none. The papers I've read haven't mentioned such. Thank you sir for informing me of such concept. I'm pretty sure it would come in handy. Now trying to search for something in that regard. – Bobby Mar 14 '14 at 06:34

0 Answers0