WISS

From IntRoLab

Jump to: navigation, search

Description

Artificial audition recently became popular in mobile robotics in order to enhance the human-robot interaction. Speech recognition is the main field of interest whereas speaker recognition receives little attention. The ManyEars system (based on the AUDIBLE project) allows a mobile robot to localize, track and separate multiple simultaneous sound sources. This system uses an array of eight microphones disposed in a cubic shape. This speaker recognition system, named WISS (Who IS Speaking), is coupled to the ManyEars system. This speaker recognition system is robust to noise and dynamic environments. Parallel model combination (PMC) and masks are used to increase the identification rate within a noisy environment. A confidence value is also introduced to weight the obtained identifications. The simplicity of this system makes it suitable for real-time applications on a General Purpose Processor (GPP).


ManyEars Overview v2.png

Fig. 1. Blocks diagram of the ManyEars system




Wiss entrainement v4.png

Fig. 2. Blocks diagram of the WISS system for training




Wiss evaluation v3.png

Fig. 3. Blocks diagram of the WISS system for identification




Team



Source code


This project is open source and available for the Matlab/Octave environments on GitHub.


Videos


Publications

  1. Grondin, F., Michaud, F. (2012), "WISS, a Speaker Identification System for Mobile Robots," Proceedings of the International Conference on Robotics and Automation: 1817-1822 (pdf)
  2. Grondin, F., Reconnaissance de locuteurs pour robot mobile, Mémoire de maîtrise, Département de génie électrique et de génie informatique, Université de Sherbrooke. (pdf)