ABSTRACT

In this paper we propose a unique approach to text dependent speaker identification using transformation techniques such as DCT (Discrete Cosine Transform) and WHT (Walsh and Hadamard Transform). The feature vectors for identification are extracted using two different techniques using the transforms, one without overlap and the other with overlap. The results show that accuracy increases as the feature vector size is increased from 64 onwards. But for feature vector size of more than 512 the accuracy again starts decreasing. The maximum accuracy without overlap is more than with overlap for both the transforms. Also the results show that DCT performs better than WHT. The maximum accuracy obtained for DCT is 94.28% for a feature vector size of 512.