The hearing system, even in front of complex auditory scenes and in unfavourable conditions, is able to separate and recognize auditory events accurately. A great deal of effort has gone into the understanding of how, after having captured the acoustical data, the human auditory system processes them. The aim of this work is the digital implementation of the decomposition of a complex sound in separate parts as it would appear to a listener. This operation is called signal separation. In this work, the separation of speech signal from complex auditory scenes has been studied and an experimentation of the techniques that address this problem has been done.

Download

Ideal three-dimensional resonators are “labeled” (identified) by infinite sequences of resonance modes, whose distribution depends on the resonator shape. We are investigating the ability of human beings to recognize these shapes by auditory spectral cues. Rather than focusing on a precise simulation of the resonator, we want to understand if the recognition takes place using simplified “cartoon” models, just providing the first resonances that identify a shape. In fact, such models can be easily translated into efficient algorithms for real-time sound synthesis in contexts of human-machine interaction, where the resonator shape and other rendering parameters can be interactively manipulated. This paper describes the method we have followed to come up with an application that, executed in real-time, can be used in listening tests of shape recognition and together with human-computer interfaces.

Download