Development of Robot Audition based on Computational Auditory Scene Analysis

计算听觉场景分析的三个主要功能，声源定位，声源分离，以及分离声音的识别，已经被开发出来，它们的集合作为开放源代码的机器人试听软件“HARK”可用。作为这次机器人试镜中概念的证明，我们开发了可以同时听说话者说话的“Shotoku王子”机器人，以及接受用户强插发声的口语对话系统。我们还开发了各种技术来分离乐器部件以进行复调演奏，并开发了实时乐谱跟踪系统。这些与音乐相关的技术被应用于使音乐机器人与人类演奏者一起演奏

Three main features of Computational Auditory Scene Analysis, sound source localization, sound source separation, and recognition of separated sounds, have been developed and their collections are made available as an open-sourced robot audition software called "HARK". As a proof of concepts in this robot audition, we developed "Prince Shotoku" robots that can listen to simultaneous talkers, and a spoken dialogue system that accepts a barge-in utterance of the user. We also developed various technologies to separate musical instrument parts for polyphonic performance, and real-time score following systems. These musical-related technologies are applied to make musical robots to play ensemble with human players