神经元网络语音增强方法及DSP实现

Neural Network algorithm for Speech Enhancement And Its Implementation

在噪声环境下双麦克风语音增强应用中,由于麦克风之间存在交叉串扰,传统自适应对消算法降噪性能受到极大的影响。为了提高降噪系统性能,提出了一种基于神经元网络双麦克风自适应抗交叉串扰语音增强方法。该方法通过设置两级自适应算法,消除了麦克风之间的交叉串扰,其中自适应算法均采用神经元网络对消方法。算法仿真基础上,运用DSP制作了实时降噪处理样机。测试结果表明,采用新方法后,系统噪声抑制性能得到了很大的提高。

The performance of traditional adaptive noise cancellation algorithms is greatly suffered by cross-talk signals in two microphone speech enhancement applications. To promote the performance of two-microphone noise cancellation algorithms, a neural net- work based system is presented to extract desired speech signal from cross-talk signals. This system consists of a two-stage adaptive filter, which can used to removal cross-talk, and neural network is adopted to realize the adaptive filters. Using DSP,We made a real time processing sample based on this new method, The real noisy environment tests show its excellent performance in noise cancellation contrast to traditional adaptive method.