Abstract
In the previous paper, we proposed an effective spectrum analysis method for transcription, in which adaptive technique was used. Frequencies analyzed by the FFT method are integral multiples of one particular frequency, but in this proposed method frequencies can be selected freely, so analysis along the musical scale without frequency correction is possible.
In this method, LMS adaptive algorithm is used. The step size parameter is a convergence factor that controls stability and adaptation rate. In this paper, we derived a method to decide the step size parameter from the condition for convergence of the coefficient and the learning curve, and from the relation between the time constants and the duration time of sound. Simulated and experimented results are illustrated.
