Neural Network algorithm for Speech Enhancement And Its Implementation
Ouyang Shan · Microcomputer Information · 2008
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 network 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.