基于加权预测误差的低复杂度去混响

Low Complexity Dereverberation Based on Weighted Prediction Error

在音视频会议以及人机交互等应用场景下,设备所采集到的音频信号往往会受到室内混响的干扰,从而降低语音的清晰度与可懂度。基于加权预测误差的自适应去混响算法是目前较为主流的盲去混响算法,该算法能够实时有效地去除混响,然而往往具有较高的计算复杂度。为降低算法的复杂度,通过分块对角矩阵简化原算法中相关的矩阵运算,实验证明,所设计的算法在确保语音质量的同时,降低了原算法的计算开销。

In application scenarios such as audio and video conferences and human-computer interaction,the audio signals collected by the device are often disturbed by indoor reverberation,thereby reducing the clarity and intelligibility of speech.The adaptive dereverberation algorithm based on the weighted prediction error is the mainstream blind dereverberation algorithm at present.This algorithm can effectively remove the reverberation in real time,but it often has high computational complexity.In order to reduce the complexity of the algorithm,the relevant matrix operations in the original algorithm is simplified by dividing the diagonal matrix.The experiment proves that the algorithm designed reduces the computational cost of the original algorithm while ensuring the voice quality.