基于卷积神经网络的面罩语音识别

针对带噪面罩语音识别率低的问题,结合语音增强算法,对面罩语音进行噪声抑制处理,提高信噪比,在语音增强中提出了一种改进的维纳滤波法,通过谱熵法检测有话帧和无话帧来更新噪声功率谱,同时引入参数控制增益函数;提取面罩语音信号的Mel频率倒谱系数(MFCC)作为特征参数;通过卷积神经网络(CNN)进行训练和识别,并在每个池化层后经局部响应归一化(LRN)进行优化.实验结果表明:该识别系统能够在很大程度上提高带噪面罩语音的识别率.

Mask speech recognition based on convolutional neural network

Aiming at problem of low mask speech recognition rate of noise mask speech,mask speech recognition method based on improved convolutional neural network (CNN ) is proposed. Combine speech enhancement algorithm to suppress noise of mask speech and increase signal-to-noise ratio(SNR). An improved Wiener filtering algorithm is proposed for speech enhancement. Using spectral entropy algorithm to detect the frame which has speech to update noise power spectrum. At the same time,introducing parameter β to control gain function. Extract Mel frequency cepstrum coefficient(MFCC) as characteristic parameters. Use CNN for training and recognization. The CNN includes two convolution layers,two pooling layers,one fully connected layer and a softmax classifier. And add local response normalization(LRN) after every pooling layer to optimize CNN. The experimental results show that the recognition system can greatly improve the recognition rate of noisy mask speech.