Navigating variation: Composing for audio Mosaicing

We present a method, applicable to corpus-based concatenative synthesis and specifically to audio mosaicing, that assists the composer in exploring the relationship between the parameterization of a concatenative algorithm and the resulting similarity between the output sound and the original target soundfile. Rather than focus solely on straightforward imitation, our work is predicated upon the notion that similarity can be manifest in a variety of perceptually meaningful ways and that both semblance and dissemblance have compositional utility. Our method consists of visualizing a collection of concatenated outputs, each of which is a unique solution to the problem of matching the same target soundfile with the same sound database but using a different combination of descriptor weights. We create a solution space where the location of each output is modeled by its similarity to the target as well as its similarity to each other solution. Visualization and navigation of this space is made possible through a multi-dimensional scaling algorithm, permitting 2D browsing, aural feedback, and the composition of paths through the solution space. This meta-control framework helps to give the composer a more comprehensive understanding of concatenative potential. By arranging concatenated outputs into different regions of similarity and dissimilarity, the solution space provides a rich and expansive terrain for compositional exploration and discovery.