This study presents a framework for audio and music processing, which comprises an analysis and a synthesis path that are connected at three-representational levels. Auditory signal analysis techniques include a multipitch analysis model, an event detector, and sinusoidal modeling which are combined in an iterative sound separation system. Techniques are presented for detection of perceptually relevant features, such as inharmonicity, vibrato and decay characteristic, from polyphonic mixtures of harmonic sounds. The integration of the analysis and synthesis parts is illustrated with examples, where two-voice acoustic guitar signals are analyzed into an object-based representation and resynthesized using sound source models.:
This paper costs $33 for non-members and is free for AES members and E-Library subscribers.