Time‐Frequency Processing: Spectral Properties
Tuomas I. Virtanen, Emmanuel Vincent, Sharon Gannot · 2018
Audio source separation and speech enhancement algorithms typically do not operate on raw time-domain audio signals, but on time-frequency representations. In this chapter we introduce the most common time-frequency representations used. We first describe the procedure for calculating a time-frequency representation and converting it back to the time domain, using the short-time Fourier transform as an example. We also present other common time-frequency representations and their relevance for separation and enhancement. We then discuss the properties of sound sources in the time-frequency domain, including sparsity, disjointness, and more complex structures such as harmonicity. Finally, we explain how to achieve separation or enhancement by time-varying filtering in the time-frequency domain. We provide tractable approximations of the exact filtering process, including the classical narrowband approximation.