Singing Transcription Technology Based On Tonal Analysis

All singing segments may contain two kinds of melodies:the actual melody and the expected melody.The difference between the two types is that the former takes the absolute pitch of each singing note as the pitch of the singing melody note,and the latter sees scale degrees as the pitch of the note.Here we will consider the expected melody as the correct melody of singing segments,we propose a singing note analysis method based on singing melody tonality analysis,design and realize an automatic singing transcription system and an evaluation method of the system’ s performance.The tonality of the best matching singing segments is estimated by analyzing intervals of neighboring or“alternating" singing notes.A new evaluation method of the singing melody estimation by comparing estimation of the traditional actual melody with that of the expected melody is established.The new evaluation method involves the correct deduction interval rate that describes accuracy of singing melody lines and the correct correction interval rate that is related to the automatic correction capability of the system.