Mel Frequency Cepstral Coefficient (MFCC) extraction for speaker identification on FPGA

Abstract

Feature extraction of speech is one of the most important issues in the field of speech recognition and representative of the speech. Mel Frequency Cepstral Coefficient (MFCC) is one the most important features required among various kinds of speech application. In this paper, FPGA-based for speech features extraction MFCC algorithm is proposed. The complexities of computational as well as requirement of memory usage are characterized, analyzed, and improved enormously. Look-up table scheme is used to deal with the elementary function value in the MFCC algorithm and fixed-point arithmetic is implemented to reduce the cost under accuracy study. The final feature extraction design is implemented effectively into the FPGA-Xilinx Virtex2 XC2V6000 FF1157-4 platform.