Performance Evaluation of Wiener Filter and Kalman Filter Combined with Spectral Subtraction in Speaker Verification System
Utpal Bhattacharjeeand, Pranab Das · 2013
This paper investigates the performance of speaker verification system in mobile environment and the techniques used to improve the robustness of the verification system. The paper demonstrates by corrupting the speech signal with additive white Gaussian noise in simulated environment. A comparative study of the three front-end noise reduction techniques namely spectral subtraction, Wiener filter and Kalman filter have been made independently as well as combining spectral subtraction with other two methods alternatively and their performances have been evaluated for the clean speech as well as contaminated speech with different level of white Gaussian noise. It has been observed that spectral subtraction plays an important role in reduction low power Gaussian noise whereas Kalman filter is efficient in reduction noise when noise power is high. Wiener filter improves the performance at all level of noise. No considerable performance improvement has been observed when spectral subtraction is combined with other two methods.