BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Hierarchical Bayesian models for audio and music processing - A. T
 aylan Cemgil\, CUED Signal Processing Lab.
DTSTART:20080916T131500Z
DTEND:20080916T141500Z
UID:TALK13417@talks.cam.ac.uk
CONTACT:Taylan Cemgil
DESCRIPTION:In recent years\, there has been an increasing interest in sta
 tistical approaches and tools from Bayesian statistics and machine learnin
 g for the analysis of audio and music signals\, driven partially by applic
 ations in music information retrieval\, computer aided music education and
  interactive music performance systems. The application of statistical tec
 hniques is quite natural: acoustical time series can be conveniently model
 led using hierarchical signal models by incorporating prior knowledge. Onc
 e a realistic hierarchical model is constructed\, many audio processing ta
 sks such as coding\, restoration\, transcription\, separation\, identifica
 tion or resynthesis can be formulated\nconsistently as Bayesian posterior 
 inference problems.\n\nIn this talk\, I will review our recent work in var
 ious signal models for audio and music signal analysis. In particular\, fa
 ctorial switching state space models\, Gamma-Markov random fields and Nonn
 egative\nmatrix factorisation models will be discussed.\nSome models admit
  exact inference\, otherwise efficient algorithms based on Monte Carlo or 
 variational approximation methods can be developed. I will \nillustrate ap
 plications in music transcription\, restoration and audio source separatio
 n.\n
LOCATION:LR12\, Engineering\, Department of
END:VEVENT
END:VCALENDAR