Farshad Jafari

Vocal Melody Construction for Persian Lyrics

2020

Deep Learning Model

Researcher, Developer

This project marked the beginning of my journey in music technology. As a writer, I’ve always been captivated by the musicality of language, a technique often present in my work. For my master’s thesis, I explored whether lyrics could have inherent musical qualities that can be modeled through music datasets. I created a dataset of Persian song lyrics and melodies and trained an LSTM network to generate new melodies based on given lyrics.

The model was tested on unseen lyrics, and the generated excerpts were performed by an expert pianist and singer. A survey conducted on the generated music showed promising results, and I documented the process and findings in a paper.

Performed Samples

GitHub page

Paper