Sixth European Conference on Speech Communication and Technology
(EUROSPEECH'99)

Budapest, Hungary
September 5-9, 1999

EasyTalk: A Large-Vocabulary Speaker-Independent Chinese Dictation Machine

Fang Zheng, Zhanjiang Song, Mingxing Xu, Jian Wu, Yinfei Huang, Wenhu Wu, Cheng Bi (1)

Speech Laboratory, Computer Science, Tsinghua University, Beijing, China
(1) Zhongshan Keysun Information Technology Co. Ltd., Guangdong, China

The EasyTalk application is a large-vocabulary speaker-independent continuous Chinese speech recognition system, i.e. Chinese dictation machine (CDM), under the WINTEL environment. Addressed in this paper are a number of novel techniques adopted in the CDM engine which is the basis of EasyTalk, including the merging-based syllable detection automaton (MBSDA) and the statistical knowledge based frame synchronous search (SKB-FSS) algorithms in the acoustic processing stage, the percentage in critical area (CAP) and recognition score gap (RSG) methods for the acceptation and rejection decision, the word search tree (WST), the N-Gram, and the syllable synchronous network search (SSNS) algorithm in the language processing stage, the embedded multiple model sheme (EMM) and the fuzzy syllable set (FSS) for the robustness purpose.


Full Paper (PDF)   Gnu-Zipped Postscript

Bibliographic reference.  Zheng, Fang / Song, Zhanjiang / Xu, Mingxing / Wu, Jian / Huang, Yinfei / Wu, Wenhu / Bi, Cheng (1999): "Easytalk: a large-vocabulary speaker-independent Chinese dictation machine", In EUROSPEECH'99, 819-822.