A time scale modification with large and varying scaling factors

Abstract

When it comes to time- and pitch-scale modification algorithms, Phase Vocoder based approaches are widely used. However, a problem with the standard approach is the limited scaling range from 50% to 200%. For an adaptive pitch transposition method for cochlear implants, larger scaling range values that are varying on a by-frame-basis are needed. This paper shows a solution for that problem with adjusted modification and synthesis stages. The developed algorithm has a constant analysis hop size and a varying synthesis hop size. Effectively, interpolated intermediate frames are introduced to compensate for missing signal information. These frames are spaced equidistant with the reciprocal scaling factor and have linear interpolated amplitudes. As a result, the desired characteristics could be achieved by a high quality implementation of the new method. The sound quality could be evaluated at an informal listening test.