University of Cambridge > Talks.cam > Machine Intelligence Lab Seminar > Spectral Moment Features for Robust Speech Recognition

Spectral Moment Features for Robust Speech Recognition

Add to your list(s) Download to your calendar using vCal

If you have a question about this talk, please contact Kai Yu.

In this talk I will present the SMAC front-end (Spectral Moment features Augmented by low order Cepstral coefficients). The SMAC feature vector comprises the first central spectral moment and the first two cepstral coefficients – C0, C1. The spectral moment component, which the primary one, is essentially a frequency estimate which is computed using a frequency domain Gabor filterbank (mel spaced). It captures the resonant structure of the speech spectrum, while the overall spectral shape is not adequately modeled. This is why the cepstral coefficients are added, the C0 as an energy estimate and C1 as a spectral tilt estimate. A key advantage of the spectral moment vector is that does not require a decorrelation transformation (e.g. DCT ) and hence the representation remains in the frequency 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 speech recognition tasks in comparison with MFCC and PLP .

This talk is part of the Machine Intelligence Lab Seminar series.

Tell a friend about this talk:

This talk is included in these lists:

Note that ex-directory lists are not shown.

 

© 2006-2019 Talks.cam, University of Cambridge. Contact Us | Help and Documentation | Privacy and Publicity