Department of Computer Science

University of Hong Kong

Pokfulafn Road, Hong
Kong

Many existing fundamental frequency recognition (FFR) algorithms return reliable results when the analysis window is sufficiently wide. In some applications, however, the response time, i.e., the sum of the width of the analysis window and the computation time for the FFR algorithm, must be made as short as possible. This paper studies the effect of window width on the accuracy of two FFR algorithms and describes a new algorithm with improved accuracy for narrow analysis windows. The new algorithm uses dynamic programming to match harmonics to peaks in the constant-Q transform of the signal. A modification to another FFR algorithm that enhances its performance in real time is also considered.