Improved closed set text independent speaker identification system using Gammachirp Filterbank in noisy environments

Several modern speaker recognition systems use a bank of linear filters as the primary step in performing frequency analysis of speech and extracting the acoustics parameters that permit characterizing the speaker identity. In this paper we point up the employ of novel feature set extracted from speech signal. The new skill for extracting these parameters is based on the human auditory system characteristics and relies on the Gammachirp Filterbank to imitate the cochlea frequency resolution with nonlinear resolution according to the equivalent rectangular bandwidth (ERB) scale. For evaluation a comparative study was operated with standard MFCC, and the effect of these differences using an usual HMM/GMM for text independent speaker recognition system, for noisy environments. Performances were test database contaminated with additive noise different real-environment noises were used: car noise provided by Volvo, factory noise and white noise from Noisex92 [1]. Tests were carried out at different SNR levels (-3dB, 0dB, 3dB, 6dB, 12dB).