Sound event recognition using python : application to humanoid robots with limited cpu

Inside [Aldebaran][1] company the goal of our team is to explore all the domains required for natural and efficient human-robot interaction. We mainly use python stack (scipy, numpy, scikit-learn) in our daily work as it provides a fast and efficient workflow to create and tune our robot behaviors.

In this presentation we will describe and demonstrate how we achieve real time sound event recognition on humanoids robots using python and scikit-learn. This work is funded by european project EARS in which our study case is a groom hotel robot that help customers : Two robots are waiting in a hotel lobby. They welcome people and give them relevant information. The robots detect and recognize sounds type such as: Door bell, phone ring, fire alarm, etc. New sounds could be easily learned by the robots.

The talk will be done by two human speakers followed by a live demonstration with NAO robots on the stage. Code and data will be published so everyone can experiments with sound recorded on the robots.