论文部分内容阅读
为提高语音清音、浊音和静默帧的分类准确率,提出了一种基于栈自动编码机的语音分类新方法.该方法由栈自动编码机和Softmax分类器组成的深度神经网络实现.首先,提取子带信号强度、残差信号峰值、增益、基音周期和线谱频率作为训练序列无监督训练栈自动编码机;然后,使用栈自动编码机的输出对Softmax分类器进行有监督训练;最后,有监督微调整个网络,得到最终网络参数.实验结果表明,在不同背景噪声及不同信噪比下,文中算法的分类准确率均优于传统算法的,且信噪比越低,性能优势越明显.
In order to improve the classification accuracy of voice unvoiced, voiced and silent frames, a new speech classification method based on stack automatic coder is proposed, which is implemented by a deep neural network consisting of stack auto-coder and Softmax classifier.Firstly, Sub-band signal strength, residual signal peak, gain, pitch period and line spectrum frequency as training sequence unsupervised training stack auto-coder; then, Softmax classifier is supervised with the output of stack auto-coder; finally, The results show that under different background noise and different signal-to-noise ratio, the classification accuracy of the proposed algorithm is better than that of the traditional algorithm, and the lower the SNR, the more obvious the performance advantage is.