基于差分过零峰值幅度特征的抗噪语音识别

A Robust Speech Recognition based on Zero-crossing with Peak Amplitude Model Combining Difference Information

针对多数语音识别系统在噪音环境下性能急剧下降的问题,提出了一种新的语音识别特征提取方法。该方法是建立在听觉模型的基础上,通过组合语音信号和其差分信号的上升过零率获得频率信息,通过峰值检测和非线性幅度加权来获取强度信息,二者组合在一起,得到输出语音特征,再分别用BP神经网络和HMM进行训练和识别。仿真实现了不同信噪比下不依赖人的50词的语音识别,给出了识别的结果,证明了组合差分信息的过零与峰值幅度特征具有较强的抗噪声性能。

To solve the problem that recognition rates of most speech recognition systems decrease severely in the noisy environments, this paper presents a new feature extraction method based on auditory model. The frequency information of speech signal is obtained by combining the zero-crossing intervals of speech signal and difference speech signal. The intensity information is obtained by peak detecting and nonlinear compression of amplitude. By combining them together speech features are got, which will be trained and recognized by BP neural network or HMM. And a 50 words iSolated speech recognition system is simulated in different noise ratio. The experimental results of using BP neural network and HMM are presented respectively. It is showed that new feature have better robustness than traditional methods.