Perceptually Smooth Timbral Guides by State-Space Analysis of Phase-Vocoder Parameters

Sculptor is a phase-vocoder-based package of programs that allows users to explore timbral manipulation of sound in real time. It is the product of a research program seeking ultimately to perform gestural capture by analysis of the sound a performer makes using a conventional instrument. Since the phase-vocoder output is of high dimen-sionality—typically more than 1,000 channels per analysis frame—mapping phase-vocoder output to appropriate input parameters for a synthesizer is only feasible in theory. For example, a lookup table of dimension 1,024 with only 16 points on each axis contains more than 10 1,200 entries. Even using simplex-based storage schemes (Bowler et al. 1990), it is clear that the number of dimensions must be reduced if direct gestural capture is to become practical. State-space models promise to reduce the dimen-sionality of the problem, but the coefficients they produce are not well correlated to the acoustic input in the way that a spectrum generated by a phase vo-coder is. The need arose to integrate within a real-time framework the facility to manipulate signals in both the frequency and Laplace domains. Sculptor uses state-space techniques to model the formant characteristics of a sound. It allows the user to use both spectrum-and filter-based resyn-thesis models, which can be edited and auditioned in real time. The source code of the package is freely available, allowing researchers to develop further functionality. Because the package combines and simultaneously presents the interfaces associated with filter-based and spectral methods, new algorithms can be tested in the appropriate domain.