Machine listening intelligence

The relation between signals and symbols is a central problem for acoustic signal processing. Among different kind of signals, musical signals are specific examples in which there is some information regarding the underlying symbolic structure. While an impressive amount of research has been done in this domain in the past thirty years, the symbolic processing of acoustic and musical signals is still only partially possible. The aim of this paper, grounded on our previous work [Cella, 2009], is to propose a manifesto for a generalised approach for the representation of acoustic and musical signals called machine listening intelligence (MLI), by integrating cognitive musicology insights, hand-crafted signal processing and deep learning methods in a general mathematical framework. Among existing approaches that share similarities with ours, there are the multiple viewpoint system [Conklin, 2013] and IDyOM [Pearce and Wiggins, 2013]. While comparing differences and similiarities with these approaches could be interesting, we will not do this here and we mention them only for reference.