Abstract
This paper proposes a pitch detection technique for transcription based on the extended-notch Fourier transform (E-NFT). The E-NFT can calculate Fourier coefficients with three data points per one frequency component. It can use to analyze a damping signal such as piano sound, and detect the start and end points of musical sounds to determine musical notes.
First, we introduce the E-NFT and then describe the realization of the automatic transcription system using the E-NFT. To simplify the processing of the E-NFT, we decompose musical sounds to three octave frequency bands using bandpass filters. Last we demonstrate the automatic transcription of clarinet (monophony) and piano (triplephony) sounds of an elecronic keyboard. It can be shown that the proposed method has a small amount of calculations and good time response comparing with other methods.
