【摘要】 随着线上音视频技术尤其是视频会议的需求量增加，语音增强技术引发广泛关注。该文提出了一种两阶段处理的多通道语音增强算法，首先对麦克风阵列采集到的多通道数据应用基于改进滤波求和网络的波束形成算法以抑制相干噪声，然后使用卷积长短期记忆网络去除波束形成器输出中残存的混响和噪声部分。仿真结果表明，该算法能够有效提高语音质量和可懂度。更多还原

【Abstract】 The rising demand for online audio and video technologies like video conferencing has driven considerable interest in speech enhancement. This study focuses on multi-channel speech enhancement with a two-stage processing algorithm. Firstly, the multi-channel data from a microphone array is performed using a beamforming algorithm based on an improved filter and sum network to suppress coherent noise. Then, a convolutional long short-term memory network is employed to remove the residual reverberation and noise components from the beamformer outputs. Experimental results demonstrate that the processing two-stage multi-channel speech enhancement algorithm can effectively improve speech quality and comprehensibility.更多还原