Using Confusion Networks for Speech Summarization

For extractive meeting summarization, previous studies have shown performance degradation when using speech recognition transcripts because of the relatively high speech recognition errors on meeting recordings. In this paper we investigated using confusion networks to improve the summarization performance on the ASR condition under an unsupervised framework by considering more word candidates and their confidence scores. Our experimental results showed improved summarization performance using our proposed approach, with more contribution from leveraging the confidence scores. We also observed that using these rich speech recognition results can extract similar or even better summary segments than using human transcripts.

• Type: Journal
• Year: 2010
• Where: NAACL
• Authors: Shasha Xie, Yang Liu
• Keywords: Computational Linguistics, Confidence Scores, NAACL 2010, Speech Recognition, Summarization Performance