Waveform-aligned adaptive windows for spectral component tracking and noise rejection

Yang Zhao

David Gerhard

pp. 463-480

in: Mitsuko Aramaki, Olivier Derrien, Richard Kronland-Martinet, Sølvi Ystad (eds), Sound, music, and motion, Berlin, Springer, 2014

Abstract

A new Short-Time Fourier Transform (STFT) pipeline is presented which uses a two-pass method to adapt the size of the analysis window to the period of the signal, which is assumed in this case to be at least pseudo-periodic. The pipeline begins with pitch estimation, followed by upsampling, to construct a new analysis window that matches the length of a single period. This reduces or eliminates the spectral leakage problems which are typical of traditional STFT analysis techniques. The result is a discrete and accurate spectral representation that provides highly accurate location of partials from one analysis frame to the next. We have extended this method to allow noise cancellation by selecting an analysis window that contains a small whole number of complete cycles. We also present a new display method based on this pipeline which greatly improves the spectrogram through enhanced distinction among partials. Finally, validation is performed by signal restoration on 40 clips, showing the superiority of the pipeline for true periodic signals and comparability for pseudo-periodic signals.