Performance Comparison of MFCC for Text Independent Speaker Identification

In this paper, a Text-Independent Speaker Identification system is implemented. The Mel Frequency Cepstral coefficients (MFCC’s) have been used for feature extraction and Vector Quantization (VQ) technique is used to manipulate the data such that it maintains the most prominent characteristics. The extracted speech features (MFCC’s) of a speaker is quantized to a number of centroids using the K-mean algorithm. These centroids constitute the codebook of that speaker. MFCC’s are calculated in both training and testing phase. The speaker is identified according to the minimum quantization distance which is calculated between the centroids of each speaker. The performance of the Mel-Frequency Cepstrum Coefficients may be affected by the different factors like number of filters, test shot length, codebook size and the type of window. In this paper, performance comparison of MFCC for the above factors is done to find a best implementation for Text Independent Speaker Identification.