Singing Voice Separation in Mono-Channel Music

With predominant communication technology in our digital world, music has turned digital, and much research has been shifted toward digital music processing. Singing voice separation is one of the active research areas since the singing voice itself contains abundant information within, such as melody, singer's characteristic, lyrics, language, emotion, etc. This wide variety of resources is very useful for diverse areas of multimedia research, including music information retrieval (MIR), singer identification, or even karaoke systems. However, this singing voice separation, especially in mono-channel environment, is very challenging problem, whose existing methods are still impractical for real world music. Our main contribution of this work is to propose a singing voice separation algorithm based on non-negative matrix factorization (NMF) method that can effectively extract the singing voice from the mono-channel music. We demonstrate its utility and effectiveness by experiments in both quantitative evaluation and its pitch detection accuracy from the separated singing voice.