Paper Title: Performance Analysis of Isolated Speech Recognition Technique Using MFCC and Cross-Correlation
Speech signal processing has become an important mode of interaction with computer. In this paper, Mel Frequency Cepstral Coefficient (MFCC) technique has been used to process speech samples to attain the recognition. MFCC is a term which narrate the short-term power spectrum of a speech signal, depend on a linear cosine transform (FFT and DCT we have used in our work) of a log power spectrum on a nonlinear Mel scale of frequency. We have used Dynamic Time Warping algorithm and Cross Correlation algorithm to match feature vectors. We have taken five recorded reference word through “One” to “Five”. Then the feature vectors generated from this reference signals are stored in database. A test sample of any numeric in “One” to “Five” is again recorded and then the algorithm is applied to recognize the same with recorded reference voices. In our paper, the recognition techniques show different percentages of accuracy. The highest recognition accuracy we got 92% for Dynamic Time Warping algorithm with FFT transformation. The accuracy for Dynamic Time Warping algorithm using DCT is 86% which is less than DP algorithm using FFT. The average accuracy for Cross Correlation using FFT is 78% and average accuracy for DCT is only 60%.