Suppression of White and Colored Noise in Bangla Speech Using Kalman Filter (ICECE 2016)

Abstract

Abstract—We introduce adaptive Kalman filter approach for suppression of white and colored noise that corrupts Bangla speech. It is difficult to retrieve the desired information from noise corrupted speech signals. Bangla speech is frequently corrupted by white and colored noise in chaotic environment and so noise suppression is extremely necessary for smooth communication purpose. Background noise in Bangla speech may be stationary or dynamic in nature. Kalman filter is a well-known recursive filter that enhances speech signals corrupted by both static and dynamic noises. In this paper, we apply adaptive Kalman filter and show that it effectively reduces white and colored noise from corrupted Bangla speech and retrieves the information carried by the original speech. We compare the retrieved speech to the corresponding noise-free speech in terms of pitch values and analyze error to evaluate performance of the proposed filter. We take oral speech as voice record by native Bangla speakers and work with both vowels and consonants to show the feasibility of Kalman filter in both cases. Both male and female voices are analyzed to prove the effectiveness of the filter for Bangla speech irrespective of gender. Elimination of noise would provide nondisruptive communication in Bangla enhancing the efficacy of Bangla speech processing.