****************************  Special Date *****************************




             Department of Systems Engineering and Engineering Management

                                 The Chinese University of Hong Kong




Title:  Unsupervised discriminative training


Speaker:  Dr. Lan Wang

                Machine Intelligence Laboratory

                Cambridge University Engineering Department(CUED)                      

Date     :   November 27th, 2006 (Monday)

Time    :   4:30p.m. - 5:30p.m.

Venue  :   Room 513

                MMW Engineering Building(Engineering Building Complex Phase 2)







A significant cost in obtaining acoustic training data is the

generation of accurate transcriptions. For some sources, closed-caption

data is available. This allows the use of lightly-supervised

training techniques. However, for some sources and languages,

closed-caption is not available. In these cases, unsupervised training

techniques must be used. This work examines the use of unsupervised

discriminative training. In unsupervised training, automatic

transcriptions from a recognition system are used fir training.

As these transcriptions may be erroneous, data selection may be useful.

Two forms of selection are described, one to remove non-target

language shows, the other to remove segments with low confidence.

Experiments were carried out on a Mandarin transcription task.

Results show that the gains from unsupervised training are highly

dependent on the accuracy of the automatic transcription.






Dr. Lan Wang is a Research Associate in the Machine

Intelligence Laboratory, Cambridge University Engineering Department

(CUED). She is working on the Autonomous Global Integrated Language

Exploitation project funded under DARPA's Global Autonomous Language

Exploitation program. She got her master degree from Peking University

(PKU) in 1996 and worked in PKU as a lecture for five years. From October

2001, she worked on Ph.D in Computer Speech and Language Processing at

CUED. She completed her Ph.D thesis on discriminative adaptive training and

adaptation in 2005 under the supervision of Prof. Phil. Woodland. Her

interests in speech research include speech recognition and biometrics



***********************  ALL ARE WELCOME  ************************


Enquiries: Peixiang Zhao or Jeffrey Xu Yu,

                 Department of Systems Engineering and Engineering Management


Website:   http://www.se.cuhk.edu.hk/~seg5810

Email:       seg5810@se.cuhk.edu.hk