This paper introduces a new extension to the concept of Audio Mosaicing, a process by which a set of unrelated sounds are blended together to form a new audio stream of shared sonic characteristics. The proposed approach is based on the algorithm that underlies the well-known channel vocoder, that is, it splits the input signals into frequency bands, which are then processed individually, and then recombined to form the output. In a similar manner, our mosaicing scheme first uses filterbanks to decompose the set of input audio segments. Then, it introduces the use of Dynamic Time Warping to perform the matching process across the filterbank outputs. Following this, the re-synthesis stage includes a bank of Phase Vocoders, one for each frequency band to facilitate targeted spectral and temporal musical effects prior to recombination. Using multiple filterbanks means that this algorithm lends itself well to parallelisation and it is also shown how computational efficiencies are achieved that permit a real-time implementation.

Download