S. Nasir, I. Khokar, and U. Akram (Pakistan)
Mel Frequency Cepstral Coefficients, Filter- Banks
The computation of speaker dependent Mel Frequency Cepstral Coefficients (MFCC) are highly influenced by the number of filters, the shape of filters, the way filters are spaced and bandwidth of the filters in a filterbank. In this paper, a comparative analysis of the effectiveness of MFCC feature vectors derived from the filterbanks, with different overlapping amongst the neighboring filters, is performed. The recognition results for 11 speakers indicate the fact that the filterbank with 25% overlapping gives the optimum solution.
Important Links:
Go Back