Exploring wikipedia's category graph for query classification
AIS'11 Proceedings of the Second international conference on Autonomous and intelligent systems
Hi-index | 0.00 |
Combining the output of several speech decoders is considered to be one of the most efficient approaches to reducing the Word Error Rate (WER) in automatic speech transcription. The Recognizer Output Voting Error Reduction (ROVER) is a well known procedure for systems' combination. However, this technique's performance has reached a plateau due to the limitation of the current voting schemes. The ROVER voting algorithms proposed originally rely on the frequency of occurrences and word level confidences, which leads to randomly broken ties and poor voting outcomes due to the unreliability of the decoder's confidence scores. This paper presents a pattern-matching-based voting scheme which has shown to reduce even further the WER.