Robust voice activity detection at noisy environment using average estimate least mean square filter

Abstract

In this paper, noise gets reduced with an average estimate LMS filter in a car noise environment so robust voice activities are being detected in a car noise environment. For the noise reduction, an average estimate LMS filter, which is a method for maintaining a source feature of speech and decreasing damages on speech information, is used in a speech signal detection process to reduce noise from a polluted speech signal. An average estimator was calculated and a step size of a LMS filter was controlled with a frame measure to improve an adaptation speed. Moreover, the starting point of result speech, comparing to the previous method of using frame energy was found to be improved by 1.7% and 3.7% of an error rate and an end point error rate, respectively. © 2013 IEEE.