Singing Voice Separation from Monaural Recordings

Separating singing voice from music accompaniment has wide applications in areas such as automatic lyrics recognition and alignment, singer identification, and music information retrieval. Compared to the extensive studies of speech separation, singing voice separation has been little explored. We propose a system to separate singing voice from music accompaniment from monaural recordings. The system has three stages. The singing voice detection stage partitions and classifies an input into vocal and non-vocal portions. Then the predominant pitch detection stage detects the pitch contour of the singing voice for vocal portions. Finally the separation stage uses the detected pitch contour to group the time-frequency segments of the singing voice. Quantitative results show that the system performs well in singing voice separation.