Signal processing for segmentation of vocal and non-vocal regions in songs: A review

It has been of great importance to us to identify a singer in a song or from collection of songs. Speaker identification is being widely used in organizing, browsing and retrieving music collections, and its speech processing techniques are well established. But the inability to delete the background accompaniment fully has set a limitation in extracting singing voice features appropriately. Singing voice identification is basically done on the basis of three techniques viz. locating vocal/non-vocal segment from the song, feature extraction of the vocal segment and statistical classification. This paper attempts to review on the various techniques developed for detection of vocal/non-vocal segments and tries to find out scope of improvement to formulate a generalized vocal detection technique.