An Improved RNN-LSTM based Novel Approach for Sheet Music Generation
Mohit Dua, Rohit Kumar Yadav, Divya Mamgai, Sonali Brodiya · Procedia Computer Science · 2020
It is very well known that Sheet Music is one of the most effective medium for musicians, professional artists and amateurs to access the chords to any songs. Many systems have already been developed which generate sheet music by taking song as input. This paper presents one such system that aims to improve the accuracy of sheet music generated by previous works. Improvement is achieved by working on source separation and chord estimation modules of the previous system. The proposed work utilizes Deep Learning techniques such as Recurrent Neural Network (RNN) with Gated Recurrent Units (GRU) and Long Short Term Memory (LSTM). In source separation module, multi-layered GRU cells for implementing RNN and in chord estimation module, LSTM cells for implementing RNN are used for the implementation. In source separation module, the number of sources that it can separate are also increased to improve the accuracy of chord estimation module.