Generating Instrumental Cover from Songs

Ali Abbas Rizvi​
2020 International Journal for Research in Applied Science and Engineering Technology  
In today's digital world and the fast-growing musical instrument technology, the ability to adapt the songs to any particular instrumental sound, becomes a priceless possession for musicians. With numerous existing tools which cater to remixing original music with instruments, a tool to convert human lyrics into instrumental rhythms is a need of the hour, or music composers of background song tracks. Often, music students start with listening to popular songs, and then attempt to recreate them
more » ... hrough instruments. This paper aims at filtering out the human vocals of songs using a library Spleeter [14], and turning them into instrumental versions, creating an ideal platform for young talented music enthusiasts to explore and learn to play various instruments. The vocals are converted to covers using the Differentiable Digital Signal Processing (DDSP) library [13] and pre-trained Convolutional Neural Networks (CNNs) -Resnet101 [15], which enables direct integration of signal processing elements with deep learning techniques. Currently in the music industry, after recording original songs with lyrics, they are re-recorded for covers. The proposed model can also help music professionals and artists in turning existing songs to a particular instrumental cover to suit their need and also allows them to release various versions of their songs, without the burden of extra effort and money.
doi:10.22214/ijraset.2020.32135 fatcat:u44hbh2rbvcgzficqzucxdgkg4