Time-Scaling of Audio Signals with Multi-Scale Gabor Analysis
The phase vocoder is a standard frequency domain time-scaling technique suitable for polyphonic audio, but it generates annoying artifacts called phasiness, or loss of presence, and transient smearing, especially for high values of the time-scale parameter. In this paper, a new time-scaling algorithm for polyphonic audio signals is described. It uses a multi-scale Gabor analysis for lowfrequency content and a vocoder with phase-locking on transients for the residual signal and for high-frequency content. Compared to a phase-locking vocoder alone, our method significantly reduces both phasiness and transient smearing, especially for high values of the time-scale parameter. For time-contraction (time-scale parameters lower that one), the results seem to be more signaldependant.