| Paper: | SLP-L6.6 | ||
| Session: | Advances in LVCSR Algorithms | ||
| Time: | Wednesday, May 17, 18:10 - 18:30 | ||
| Presentation: | Lecture | ||
| Topic: | Speech and Spoken Language Processing: Lattices and Multi-pass strategies | ||
| Title: | A Comparative Study of Discriminative Methods for Reranking LVCSR N-Best Hypotheses in Domain Adaptation and Generalization | ||
| Authors: | Zhengyu Zhou, Chinese University of Hong Kong, Hong Kong SAR of China; Jianfeng Gao, Microsoft Corporation, United States; Frank K. Soong, Microsoft Research Asia, China; Helen Meng, Chinese University of Hong Kong, Hong Kong SAR of China | ||
| Abstract: | This paper is an empirical study on the performance of different discriminative approaches to reranking the N-best hypotheses output from a large vocabulary continuous speech recognizer (LVCSR). Four algorithms, namely perceptron, boosting, ranking support vector machine? (SVM) and minimum sample risk (MSR), are compared in terms of domain adaptation, generalization and time efficiency. In our experiments on Mandarin dictation speech, we found that for domain adaptation, perceptron performs the best; for generalization, boosting performs the best. The best result on a domain-specific test set is achieved by the perceptron algorithm. A relative character error rate (CER) reduction of 11% over the baseline was obtained. The best result on a general test set is 3.4% CER reduction over the baseline, achieved by the boosting algorithm. | ||
©2018 Conference Management Services, Inc. -||- email: webmaster@icassp2006.org -||- Last updated Friday, August 17, 2012