THE SLIDING PHASE VOCODER

The Sliding Discrete Fourier Transform (Sliding DFT) is used as the engine of a phase vocoder, to create a Slid- ing Phase Vocoder (SPV). With a little care this allows very accurate pitch shifting and low latency, and opens a number of possible extensions. We also consider the use of vector parallel processing to make these techniques a viable option. 1. BACKGROUND In his AES lecture, Moorer(11) considered what might be the state of audio processing in 2020, given the ongoing increase in computer power as defined by Moore's Law; indeed, with the increasing availability of multiple-core devices at consumer prices, it seems likely that if any- thing, the trend defined by Moore will accelerate rather than decline. His predictions included the routine use of frequency-domain processing of audio where he suggested that the Sliding DFT (SDFT), in which the analysis frame is updated every sample, might offer some advantages. In (3) we described a working implementation of the SDFT, with some initial qualitative assessments that indicated t hat the SDFT merited further investigation. Accordingly, we only give a brief description in section 2. In this paper we consider the properties of the SDFT and its application to the conventional phase vocoder (pvoc). The SDFT is computationally expensive despite its ad- vantages, but multiple processor cores on one chip changes the rules; we are also working with ClearSpeed Technol- ogy plc(12) on deploying their vector processor hardware in the general area of High-Performance Audio Comput- ing (HiPAC), and the SDFT is a clear target for this work.