Abstract
    The audio signal by extracting feature parameters, using GMM as a classifier, ultimately audio classification and obtain a higher classification accuracy. Audio characteristic signal recognition method for solving general use pattern matching. First extract audio features through pretreatment, as a model of the audio clip. The model with the known reference model is compared to obtain the best matching reference model as a recognition result, research in the paper is based on simulation results of traditional BP neural network technology audio audio classification and GMM classification comparison, Compare the advantages and disadvantages of both, thus familiar with both technology, the basic working principle and algorithms. Its performance through subjective evaluation experiments and objective data analysis, for the selected audio signal BP training network has a faster speed, but GMM technique has a high recognition rate. BP network learning and memory have instability. In other words, if you increase the study sample, the trained network will need to start training from scratch, for the previous weights and thresholds are no memory. But it can be predicted, classification or clustering do better to save weight.