Abstract
This paper describes a technique of a beat tracking that recognizes temporal positions of A cappella song. The technique is divided into four procedure; acquisition of data, FFT analysis, temporal-positions recognition and rhythm recognition. The temporal-positions recognition utilizes that frequency spectra change suddenly at temporal positions. The rhythm recognition is based on the eight note, and finds out the tempo of a song when the error margin at a theoretical and an obtained temporal position is small. It was found from the experimental result that the technique is able to recognize the rhythm of the voice synthesis song by means of performing FFT analysis to 256 data every 200 or 300 times.
