Modifying STFT bins creates edge transients that synthesis windows imperfectly suppress

Evidence

When the STFT data are arbitrarily modified, the time-domain representation of the same data no longer has the original windowed shape. Furthermore, due to the circular representation, the values at the edges of the frame may deviate from zero, as shown in Figure 15.8. The non-zero edges then produce wideband transient artefacts in the output signal, and the conventional method for suppressing such effects is to apply a synthesis window, as shown in Figure 15.8. The synthesis window forces the time-domain signal values to zero at the start and at the end of the frame. The suppressed components of the signal are basically lost, which again may cause quality degradation. β€” Ville Pulkki, Matti Karjalainen, Communication Acoustics- An…, 2108

Categorization

  • XREF: Explains the classic STFT flickering/aliasing tradeoff against time-frequency resolution that connects to phase vocoder and audio effect literature.