An optimal filter for extracting time-varying harmonics such as a voiced speech or vibration sound from the noise-corrupted observation is presented. The Wiener filter analysis to the quasi-periodic signal involving random amplitude/pitch fluctuations leads to the constant-BW/constant-Q mixed comb filter in which each pass-band located at harmonic frequencies has a constant bandwidth in lower frequency, constant-Q characteristic in higher frequency, and those mixture appears in the intermediate frequency. Some numerical examples show that the frequency response fits well the filter bank structure built in our auditory system and advantageous to separating a voiced speech from background noise
J-STAGEがリニューアルされました! https://www.jstage.jst.go.jp/browse/-char/ja/