Soundscape Generation System
We developed a generative system that aims at simplifying the authoring process, but offering at the same time a realistic and interactive soundscape. A sample-based synthesis algorithm is driven by graph models (see figure). Sound samples can be retrieved from a user-contributed audio repository. The synthesis engine runs on a server that gets position update messages and the soundscape is delivered to the client application as a web stream. The system provides standard format for soundscape composition.The system includes an authoring module to create the soundscapes, and the actual audio generation engine. All code has been implemented in Supercollider, and is available under the GNU-GPL license. The system has dependencies on other SuperCollider packages (GeoGraphy, XML).
Sound Texture Synthesis
We are working on new parametric and non-parametric statistical models for synthesizing environmental sound textures, such as running water, rain, and fire. Sound texture analysis is cast in the framework of multiresolution statistical models. We stochastically sample from a model that has been trained on source sounds and captures correlations of these sound textures in a transformed feature space representation. By reconstructing a time-domain signal from the sampled feature sequences, e.g. by the inverse wavelet transform or corpus-based synthesis, we aim to create distinct but perceptually similar versions of a sound. Sound Examples