An open source software system for robot audition HARK and its evaluation

摘要

Robot capability of listening to several things at once by its own ears, that is, robot audition, is important in improving human-robot interaction. The critical issue in robot audition is real-time processing in noisy environments with high flexibility to support various kinds of robots and hardware configurations. This paper presents open-source robot audition software, called “HARK”, which includes sound source localization, separation, and automatic speech recognition (ASR). Since separated sounds suffer from spectral distortion due to separation, HARK generates a temporal-frequency map of reliability, called “missing feature mask”, for features of separated sounds. Then separated sounds are recognized by the Missing-Feature Theory (MFT) based ASR with missing feature masks. HARK is implemented on the middleware called “FlowDesigner” to share intermediate audio data, which provides real-time processing. HARK’s performance in recognition of noisy/simultaneous speech is shown by using three humanoid robots, Honda ASIMO, SIG2 and Robovie with different microphone layouts.