A Singing Voice Detection Algorithm Based on Deep Residual Convolutional Neural Network

Singing voice detection is an important segment in the field of musical artificial intelligence, and it is also a necessary technology or enhancement technology for various related studies, In this paper, we propose an algorithm based on deep residual convolutional neural net-work, The multi level convolutional neural network can learn more valid singing features thana shallow convolutional neural network when only simple and plain features are input, thereby improving the overall performance of the algorithm. In this paper, based on two basic residual network structures, six convolutional neural networks with different depths are designed. By comparing with the experimental results of the baseline system, it is proved that the performanee of the algorithm in this paper is ahead of the algorithm based on shallow convolutional neural network. At the same time, the network depth adjustability of this algorithm also adds more flexibility to its application in practice.

In Journal of Jinling Institute of Technology