Renica based music source separation for automatic Music emotion classification
Rosli Nurlaila, Nordiana Rajaee, David B. L. Bong · Unimas Institutional Repository (Universiti Malaysia Sarawak) · 2018
In music source separation, we deal with the problem of precisely separating the singing voice and instrumental accompaniment estimation in music mixtures.This problem is addressed through the use of signal processing algorithm called RENICA which refers to the combination of source separation methods namely, REpeating Pattern Extraction Technique (REPET), Nonnegative Matrix Factorization (NMF) and Independent Component Analysis (ICA).The separated estimation is later used for parameters modelling in automatic Music Emotion Classification (MEC).This paper aims to improve singing voice and instrumental accompaniment separation in 120-180 sec music signal length by merging three music source separation algorithms.From the experimental results obtained, this new combination of algorithms not only succeeds in separating better estimation but enhances the accuracy for emotion based music classification up to 97% for angry, peaceful, happy and sad emotion categories.