RetroSpat: a Perception-Based System for Semi-Automatic Diffusion of Acousmatic Music

We present the RetroSpat system for the semi- automatic diffusion of acousmatic music. This system is intended to be a spatializer with perceptive feedback. More precisely, RetroSpat can guess the positions of physical sound sources (e.g. loudspeakers) from binaural inputs, and can then output multichannel signals to the loudspeak- ers while controlling the spatial location of virtual sound sources. Together with a realistic binaural spatializatio n technique taking into account both the azimuth and the distance, we propose a precise localization method which estimates the azimuth from the interaural cues and the distance from the brightness. This localization can be used by the system to adapt to the room acoustics and to the loudspeaker configuration. We propose a simplified sinu- soidal model for the interaural cues, the model parameters being derived from the CIPIC HRTF database. We extend the binaural spatialization to a multi-source and multi- loudspeaker spatialization system based on a static adap- tation matrix. The methods are currently implemented in a real-time free software. Musical experiments are conducted at the SCRIME, Bordeaux.