Tibetan Voice Activity Detection Based on One-Dimensional Convolutional Neural Network
Gaopan Li, Rangzhuoma Cai, Cai Zhijie, Dan Chen · 2021
In order to improve the accuracy and robustness of voice activity detection in complex noise environment, this paper proposes a voice activity detection method based on one-dimensional convolutional neural network and applies it to Tibetan. Aiming at the one-dimensional feature and time-varying nature of the speech signal, the input layer, convolution layer and pooling layer of the traditional convolutional neural network are changed from two-dimensional structure to one-dimensional. While simplifying the neural network structure, it realizes accurate detection of Tibetan voice activity in complex noise. The experimental results show that compared with the voice activity detection method based on two-dimensional convolutional neural network, the voice activity detection method proposed in this paper is more accurate and robust.