基于循环神经网络的歌曲旋律与人声分离系统设计与实现

Design and Implementation of Singing-Separation System Based on Recurrent Neural Network

为了解决现今计算机市场缺乏分离歌曲人声与伴奏软件的现状,通过分析当下国内外两种主流的人声分离算法,充分总结其优势和不足,并将人声分离的思想与深度学习算法相结合,提出利用循环神经网络处理歌曲进行分离的构想,并通过实践获得成功。同时将其与网络技术相结合,建立可供用户实时将本地歌曲在线分离的交互性网页系统,弥补相关需求的空缺。

In order to solve the problem that current computer market does not have a software for users to do singing separation,by analyzing two kinds of main algorithms of singing separation at home and abroad as well as thoroughly summing up what their theory are good at and weak of,proposes an idea about using Recurrent Neural Network to do song separation and achieve success in practice.Combines this thought with network technology and establishes an interactive Web system for users to upload their songs and do separations online,which fills the gap in software market demand.