Transcription of sung song

This paper describes a system which can automatically transcribe sung songs into music note. Pitch detection and segmentation of sung songs are more difficult than that of instrumental tone. Because there are anbiguous transients, such as vibrato, between adjacent tone. In this paper, we use knowledges of the music, for example, tempo, musical form and key of the song. The division of the time into the sub-unit is performed in the first place by using musical form and tempo. Next the pitch of sub-unit is detected by the interpolation in the complex spectrum. The key of the song is used to correct errorsof the pitch of sub-units. In the ethnomusicology, there is a large number of folksongs which have not been transcribed yet. This system can serve as a supporting facility for the ethnomusicology.