Multi-speaker adaptation for robust speech recognition under ubiquitous environment

This paper presents a multi-speaker adaptation for robust speech recognition under ubiquitous environment. The goal is to adapt the speech recognition model for each speaker correctly in ubiquitous multi-speaker environment. We integrate speaker recognition and unsupervised speaker adaptation method to promote the speech recognition performances. Specifically we employ a confidence measure to reduce the possible negative adaptation caused by the environment noise or the recognition errors. The experimental results show that the proposed framework can efficiently promote the average recognition accuracy to 80~90% for multi-speaker ubiquitous speech recognition.