The ROUTE Project

Summary

The ROUTE project establishes a cooperation between the Analsysis/Synthesis team (IRCAM) and the ISIR Laboratory (UPMC).

The project focuses on a common issue for two a priori distinct research areas:

1. The automatic analysis of a sound scene, from a dictionary learning process. In audio signal processing, we find this aspect of major interest in several applications, computational auditory scene analysis (CASA), automatic indexing, source separation, detection and localization of sound objects, among others.
2. Artificial hearing, as a new study in robotics, for which the analysis of a sound stage gradually becomes a necessary prerequisite for any modern application (for example, for monitoring of the elderly, or the study of the interaction between humans and robots).

The central objective of the project is the design and development of a new method allowing for the detection and localization of the main speaker in the sound scene. The method is expected to enable a robot to identify a voice signal in the presence of noise, and locate the position of the main speaker. The problem is very closely linked to the field of computational auditory scene analysis (CASA) that has the objective to understand human sound perception and to design automated systems with a perception that mimics human hearing.

The ROUTE project takes a different perspective, even though the audio processing tools to be used are comparable (machine learning, source separation). The problem is treated from the point of view of the robot, and the interest is concentrated in the analysis audio scene in itself. The system to be developed is based on a detection and sound source localization algorithm for multichannel audio, which was developed by the Analysis / Synthesis team in the European project 3DTVs. The new perspective opened by this project is to confront advanced automatic analysis methods with the constraints and specificities of the field of robotics. The issues raised are then, among others, the following: what are the priorities that the robot must take into account when analyzing the sound stage, how to take advantage of the structure and mobility of the robot to enhance the detection, and what are the possibilities that allow simplifying the complexity of the algorithm such that an implementation within an autonomous robot becomes achievable.

People Involved

IRCAM: Marco Liuni, Nicolas Obin, Axel Roebel

ISIR: Jean-Luc Zarader, Sylvan Argentieri, Bruno Gas