Effect of speech coding on speaker identification

The increasing use of wireless systems is creating great deal of interest in the development of robust speech systems in wireless environment. The major degradations involved in wireless environment are: effect of varying background conditions, degradation due to speech coders and errors due to wireless channels. In this paper, we presented the effect of speech coding on text independent speaker identification (SI). Speech coders considered in this work are GSM full rate (ETSI 06.10), CELP (FS-1016), and MELP (TI 2.4 kbps). The amount of distortion introduced by coding is measured using log-likelihood ratio (LLR), weighted spectral slope (WSS) and log-spectral distance (LSD). The effect of coding on SI is analyzed by building SI system using both vocal track system and excitation source features. We observed that there is a significant reduction of performance in SI system due to coding, and effect is more prominent in case of SI system build with source features. We also observed that, speaker characteristics are well preserved in case of MELP compared to CELP even though MELP coder bit rate is less than CELP.