Robot Audition

This line of research intends to emulate in a non-human entity the human ability to listen, and apply these techniques in areas of technological, social and environmental impact, such as Service Robotics, the design of Assistive Listening Devices and Bioacoustics. It aims to develop techniques of Robot Audition with a focus on "light" solutions: using a few microphones and low computing resources. This is to increase their viability in the different areas of impact, and, at the same time, to make them relatively easy to port between them. Emulating human hearing involves a variety of concepts and development of a wide range of disciplines such as Signal Processing, Psychoacoustics, even Cognition.

Project listing:

Seguimiento de Fuentes Sonoras Móviles

Tracking of Multiple Sound Sources

An essential part of any Robot Audition system is the localization of sound sources. This localization is presented by its relative direction to the service robot. An important current challenge is to locate and track multiple simultaneous sound sources in highly reverberant and noisy environments. This project aims to address this challenge using few microphones, and, by using information from the robot position over time and acoustic environment, refine its localization estimates as well as jointly estimate the distance from the sound source to the robot.
Separación de Fuentes en Línea.

On-line Source Separation

Having located the sources, it is of interest to separate the auditory data coming from each source to reduce the complexity of consequent audio processing modules, such as Speech Recognition (ASR) or Source Classification. Similarly, the project will address this challenge by using few microphones and the developed techniques developed will be evaluated in real acoustic environments.
Clasificación de Tipo de Fuente

Clasificación de Tipo de Fuente

When the auditory data has been separated, and before feeding it to the Speech Recognizer, it is important that only the data from human users is fed, not noise sources. For this purpose, it is proposed to develop techniques for automatic classification of the type of sound source.