Parametric Audio Processing

The basic notion of parametric audio processing is to decompose speech and music signals using models that express the signal in terms of physically meaningful parameters and to process these parameters rather than the original signals. By doing this, computationally efficient processing can be achieved and abstract information can be extracted. It is the goal of this project to find parametric models that may be used for analysis, compression, and modification of music signals. The project will be focused on the identification of models and methods for finding the parameters that describe them. It is at the heart of this project to take into account prior knowledge about the origin and structure of music signals as well as the human auditory system. This will facilitate much more efficient compression and enable analysis of complicated mixtures of instruments. The methods and models will be evaluated in terms of the perceived quality of synthesized audio, the statistical properties of the estimators, and by comparing extracted information to annotated music.