采用了小波去噪的思想，多姿态，多角度，有不同光照,采用累计贡献率的方法，BP神经网络用于函数拟合与模式识别，从先验概率中采样，计算权重，仿真效率很高的，实现了对10个数字音的识别程。- Using wavelet denoising thought, Much posture, multi-angle, have different light, The method of cumulative contribution rate BP neural network function fitting and pattern recognition, Sampling　a priori probability, calculate the weight, High simulation efficiency, Realization of 10 digital audio recognition progra.