7th International Conference on Spoken Language Processing

September 16-20, 2002
Denver, Colorado, USA

Run Time Information Fusion in Speech Recognition

Chengyi Zheng, Yonghong Yan

Oregon Health & Science University, USA

Approaches have been studied to utilize the complementary information from different knowledge sources in Automatic Speech Recognition (ASR). These approaches can be classified into two categories, the pre-recognition fusion and the post-recognition fusion. A common problem of those approaches is that complementary information is exploited either before or after recognition. To avoid unrecoverable information loss due to pruning in decoding stage, and to better utilize the complementary information, we propose a duringrecognition information fusion scheme. Experimental result based on the proposed run-time fusion is reported. A significant improvement was observed.


Full Paper

Bibliographic reference.  Zheng, Chengyi / Yan, Yonghong (2002): "Run time information fusion in speech recognition", In ICSLP-2002, 1077-1080.