Automatic Speaker Recognition using Stationary Wavelet Coefficients of LP Residual
V. R. Sreehari, Leena Mary · 2018
Automatic speaker recognition is a challenging task when the duration of the speech utterances are very short ie., a few seconds. Source features extracted from speech utterances are shown to be effective for such cases. This paper proposes a system based on LP residual for text independent speaker recognition. Parameterization of LP residual is attempted here using Discrete Wavelet Transform (DWT) and Stationary Wavelet Transform (SWT). Feature obtained from different levels of decomposition are used for implementing an i-vector/PLDA based speaker recognition system. Effectiveness of the system is evaluated for 10 sec training - 10 sec testing task of NIST SRE 2010 database. Features obtained from SWT level-2 decomposition gives the lowest Equal Error Rate (EER) of 39.