Unisoner: An Interactive Interface for Derivative Chorus Creation from Various Singing Voices on the Web

This paper describes Unisoner, an interface for assisting the creation of derivative choruses in which voices of different singers singing the same song are overlapped on one common accompaniment. In the past, it was time consuming to create such choruses because creators have to manually cut and paste fragments of singing voices from different singers, and then adjust the timing and volume of every fragment. Although several interfaces for “mashing up” different songs have been proposed, no mash-up interface for creating derivative choruses by mixing singing voices for the same song has been reported. Unisoner enables users to find appropriate singers by using acoustic features and metadata of the singing voices to be mixed, assigning icons of the found singers to each phrase within a song, and adjusting the mixing volume by moving those icons. Unisoner thus enables users to easily and intuitively create derivative choruses. It is implemented with several signal processing techniques, including a novel technique that integrates F0-estimation results from many voices singing the same song to reliably estimate F0 without octave errors.