BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Bayesian Learning Approaches for Speech Recognition - Professor Je
 n-Tzung Chien (National Cheng Kung University\, Taiwan)
DTSTART:20090904T120000Z
DTEND:20090904T130000Z
UID:TALK19605@talks.cam.ac.uk
CONTACT:Dr Marcus Tomalin
DESCRIPTION:In this talk\, I will present my previous and ongoing studies 
 on Bayesian \nlearning for speech recognition. In the areas of speech reco
 gnition\, \nBayesian adaptation has been widely presented to deal with the
  issue of \nspeaker adaptation where the likelihood function of adapation 
 data and the \nprior density of the existing model are merged to find the 
 adapted model for \nnew speaker. Such a Bayesian learning approach is not 
 only useful for model \nadaptation but also for model regularization where
  the regularized hidden \nMarkov models (HMMs) are good for prediction of 
 unknown test data. The \nregularized HMMs can be applied for decision tree
  state tying in a data \ngeneration model and even can be integrated with 
 a large magin classifier to \nimprove the generalization of a discriminati
 ve model based on the large \nmargin HMMs. Furthermore\, the Bayesian lear
 ning is beneficial for topic-\nbased language model under the paradigm of 
 latent Dirichlet allocation (Blei \net al.\, 2003). A Bayesian topic-based
  language model shall be presented for \nspeech recognition. This regulari
 zed language model is established according \nto the marginal likelihood o
 ver the uncertainties of latent topics and topic \nmixtures. The topic inf
 ormation is extracted from the n-gram events and \ndirectly applied for sp
 eech recognition. At last\, I will summarize my \nviewpoints about the stu
 dies of Bayesian learning and address the other \nchallenging topics of ma
 chine learning methods for speech recognition.
LOCATION:LR5\, Engineering Department\, Baker Building
END:VEVENT
END:VCALENDAR