Influence of G.729 Coded Bitstream in Automatic Speaker Recognition Using Gaussian Mixture Model

Abstract

This paper investigates on the contribution of Line Spectral Frequency (LSF) derived directly from the G.729 encoded bitstream on Speaker Recognition System over IP network. In this work theGaussian Mixture Model (GMM) are used for the modeling task. The performance of the proposed system is analyzed from either the synthesized speech and directly from the coded parameters. The effect of the transmission channel is also studied and the channel is simulated using Binary Symmetric Channel (BSC) at different probability values. The recognition phase was tested with ARADIGIT corpus which is a database of Arabic spoken digits, spoken by Algerianspeakers. The obtained results show that the emerging method that uses the LSF derived directly from the G.729 bitstream improves significantly the recognition accuracy compared with the Mel Frequency Cepstral Coefficients (MFCC) features extracted fromsynthesized speech.

Keywords

• G.729
• MFCC
• LSF
• GMM
• BSC
• Speaker recognition