A Deep-Learning Based Framework for Source Separation, Analysis, and Synthesis of Choral Ensembles

Choral singing in the soprano, alto, tenor and bass (SATB) format is a widely practiced studied art form with significant cultural importance. Despite popularity of choral setting, it has received little attention field Music Information Retrieval. However, recent publication high-quality datasets as well developments deep learning based methodologies applied to music speech processing, have opened new avenues for research this field. In paper, we use some publicly available train evaluate state-of-the-art source separation algorithms from domains case singing. Furthermore, existing monophonic F0 estimators on separated unison stems propose an approximation perceived signal. Additionally, present set applications combining proposed methodologies, including synthesizing single singer voice unison, transposing remixing into synthetic multi-singer We finally conduct listening tests perform perceptual evaluation results obtain methodologies.