基于特征空间逼近的传声器阵列语音信号增强

对现有的基于自动波束形成的传声器阵列语音信号增强算法提出了改进,将各传声器采集到的信号利用ABF(自适应波束形成)进行延时补偿并求和,消除信号中弱相干和不相干噪声;再利用特征空间逼近的方法进一步去除残留的噪声。将一种定阶方法应用到基于特征空间分解的语音信号增强中,利用其“最大稳定”原理,使得有效信号模型的阶数尽可能不受原始信号信噪比的影响,消除了传统定阶过程中的随意性和不稳定性。仿真结果表明:把自适应波束形成技术和特征空间逼近的方法结合起来,能够取得良好的去噪效果。

A Method Based on Multi-microphone for Speech Enhancement by Eigenvalue Subspace Approach

A method using automatic beamforming (ABF) technology and eigenvalue subspace approach for speech enhancement is proposed. The signals are received by a microphone array that consisted of microphones. The method has two steps. First, automatic beamforming technology is adopted to combine signals from microphone array, then, signal subspace approach method is taken to cancel noises in the result of first step. In eigenvalue subspace approach based voice signal enhancement, the order of signal model is set as stable as possible, and unaffected by SNR of original signal, and randomization and instability that existed in traditional order-determination method are overcome. The computer simulations show that this method is effective.