System-on-Chip Design for Audio Processing

Abstract

Nowadays System-on-Chip (SoC) is present in every electronic system. SoC popularity is based on higher performance, reduced size, less power consumption, and alleviation of time to market by design reuse. Device scaling enabled SoC to integrate more functionality into a single chip and hence system complexity, like Audio Processing system, is no more barriers for the SoC designer. Speaker recognition/verification is one of the applications in biometrics for preventing identity fraud. It is suitable for real time scenarios and remote recognition over phone. In this project, I have designed a SoC system for Audio Processing on Altera DE2 board, FPGA platform, which automatically verify or recognize the speaker Identity. Mel Frequency Capestral Coefficient (MFCC) is used for feature extraction of the voice signal. Large samples of extracted feature are used to train the system by using Backpropagation Neural Network. After training, speaker verification done in real time by first extracting speaker voice feature, applying trained network on extracted feature, and comparing it with the stored database. Experimental result shows that the designed system is able to verify person’s identity.