BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Spectral Moment Features for Robust Speech Recognition - Pirros Ts
 iakoulis (Institute for Language and Speech Processing (ILSP))
DTSTART:20110531T120000Z
DTEND:20110531T130000Z
UID:TALK31505@talks.cam.ac.uk
CONTACT:Kai Yu
DESCRIPTION:In this talk I will present the SMAC front-end (Spectral Momen
 t features Augmented by low order Cepstral coefficients). The SMAC feature
  vector comprises the first central spectral moment and the first two ceps
 tral coefficients - C0\, C1. The spectral moment component\, which the pri
 mary one\, is essentially a frequency estimate which is computed using a f
 requency domain Gabor filterbank (mel spaced). It captures the resonant st
 ructure of the speech spectrum\, while the overall spectral shape is not a
 dequately modeled. This is why the cepstral coefficients are added\, the C
 0 as an energy estimate and C1 as a spectral tilt estimate. A key advantag
 e of the spectral moment vector is that does not require a decorrelation t
 ransformation (e.g. DCT) and hence the representation remains in the frequ
 ency domain. A second inherent property is that it has zero mean value. I 
 will show recognition results on the TIMIT\, Aurora 2\, and Aurora 3 speec
 h recognition tasks in comparison with MFCC and PLP. 
LOCATION:Cambridge University Engineering Department\, Lecture Room 6
END:VEVENT
END:VCALENDAR