5th International Conference on Spoken Language Processing
Our goal is to develop a singing synthesis system, in which "singing units" are automatically extracted from existing musical recordings of a singer accompanied by a musical instrument (piano). This paper concentrates on the problem of separating singer from accompaniment. Existing separation methods require the knowledge of the exact frequencies of the signals to be separated, and are prone to degrading the quality of the separated signals. In this paper, we use the framework of the sinusoidal modeling approach. We suggest the use of further sources of information, available to the specific task: advance knowledge of the music score, knowledge about the piano sound, and a relatively large database of the piano and singer signals, which is used to build a model of the piano sound. Results show that using musical score information and piano note modeling can improve separation quality.
Bibliographic reference. Meron, Yoram / Hirose, Keikichi (1998): "Separation of singing and piano sounds", In ICSLP-1998, paper 0416.