Robot Audition – Hands - Free Automatic Speech Recognition under Highly - Noisy Environemnts – Kazuhiro NAKADAI

This paper addresses robot audition, which realizes listening capabilities for robots using robot-embedded microphones. For robot audition, we propose real-time sound source separation and automatic speech recognition (ASR) techniques for dynamically changing environments based on microphone array processing, which is applicable to hands-free ASR under highly-noisy environments. Implementation of the proposed techniques is open-sourced as robot audition software called “HARK.” We show the effectiveness of these techniques through applications of HARK to robots.