Speech Dereverberation Using 2D Microphone Array

Nowadays, smart speech devices have been widely used for human-computer interaction. However, in many environments, the reverberation caused by the multi-path effect can seriously affect the recognition performance of speech devices. Therefore, it is crucial to eliminate reverberation in received speech signals. In this paper, we propose WPE-Beamforming, an iterative framework that can effectively perform dereverberation based on a 2D microphone array. Specifically, we first estimate the Angle of Arrival (AoA) of the sound source based on the improved 2D Multiple Signal Classification (MUSIC) algorithm. Then, using the estimated AoA, we calculate the Time Difference of Arrival (TDOA) between each microphone and use TDOA to compensate for the prediction delay in the conventional Weighted Prediction Error (WPE) algorithm. Finally, we integrate the improved WPE algorithm with the adaptive beamforming into an iterative framework to eliminate the impact of noise on the dereverberation process. Experimental results show that our proposed method achieves high scores on multiple evaluation metrics, and the word error rate of the automatic speech recognition system is only 9.89%.