BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:The effect of normalization -- a case study in speech synthesis - 
 Matt Shannon (University of Cambridge)
DTSTART:20110303T140000Z
DTEND:20110303T153000Z
UID:TALK29124@talks.cam.ac.uk
CONTACT:Shakir Mohamed
DESCRIPTION:Undirected graphical models are ubiquitous in application doma
 ins of machine learning.  However the normalization constants in these mod
 els are often difficult to compute\, and as a result are frequently droppe
 d altogether.  In this talk we'll look at the qualitative effect of this l
 ack of normalization in the domain of statistical speech synthesis. \n\nSp
 ecifically we'll compare the predictive distributions of the standard unno
 rmalized speech synthesis model\, its globally-normalized undirected count
 erpart\, and a more tractable directed graphical model.  Along the way\, w
 e'll highlight some of the general issues surrounding the choice between u
 ndirected and directed graphical models for sequence data.\n\nThe introduc
 tion to speech synthesis I will give will be aimed entirely at machine lea
 rners with no background in modelling speech\, and will I hope be realisti
 c (close to state-of-the-art)\, self-contained and framed in terms of\npro
 babilistic modelling.\n
LOCATION:Engineering Department\, CBL Room 438
END:VEVENT
END:VCALENDAR