Objective and subjective evaluation of complementary Wiener filter for speech dereverberation

Abstract

Acoustic distortion caused by reverberation can degrade speech quality and performance of speech-based systems. Several dereverberation techniques have been proposed in the literature. For example, a dereverberation method using a complementary Wiener filter can suppress late reverberation with few computational resources. As a method for dereverberation, the method using a complementary Wiener filter has been proposed, and for the exponential decay impulse response model, it is shown theoretically that we can suppress reverberation with few computational resources. In this report, we approximate expectation of the power spectrum. which is necessary to calculate a complementary Wiener filter as exponential moving average. We conducted dereverberation experiments using actual environment room impulse response. The results of the objective evaluation show that the suppression performances of the actual environment room impulse response can approximate from the results of the exponential decay impulse response model. Additionally, we investigated the relationship between the results of objective evaluation and the results of subjective evaluation. In a small reverberation environment, we can see strong correlation between the results of objective and subjective evaluation.