【摘要】 通过将ANN(人工神经网络)改进应用到HMM(隐马尔科夫模型),使用Mel频率倒谱系数(MFCC)+帧能量+MFCC一阶差分,二阶差分的结构提取咳嗽音特征参数,HMM输出的所有状态累积概率作为ANN的输入序列进行非线性映射,进而提取新的信息来提高HMM的识别性能。实验证明,利用HMM-ANN混和模型来处理咳嗽声识别具有更高的识别精度和可靠性。更多还原

【Abstract】 By applying the improved ANN(Artificial Neural Network) to HMM(Hidden Markov Model),the characteristic parameters is extracted from cough with the structure of frame energy + MFCC + its first-order and second-order difference parameters.All of the state cumulative probability of HMM outputs will server to be the input sequences of ANN for nonlinear mapping.Then the extraction of new information is conducted to improve the recognizing performance.Experimental results show that compared with the traditional single model,HMM-ANN hybrid model has a higher recognizing accuracy and reliability in dealing with cough identification.更多还原