Odyssey 2012 - The Speaker and Language Recognition Workshop

Singapore
June 25-28, 2012

On the use of Asymmetric-shaped Tapers for Speaker Verification using i-Vectors

Md Jahangir Alam (1,2), Patrick Kenny (2), Douglas O'Shaughnessy (1)

(1) INRS-EMT, University of Quebec, Montreal, Canada
(2) CRIM, Montreal, Canada

This paper presents asymmetric-shaped tapers (or windows) for speaker recognition. Symmetric tapers (e.g., hamming), having the linear phase property and longer time delay, are widely used for short-time analysis of speech signals. Since human speech perception is relatively insensitive to short-time phase distortion, the linearity constraint on phase can be removed without any adverse effects. Use of asymmetric tapers, having better magnitude response and shorter time delay, in speaker recognition can lead to a better recognition performance. Speaker verification results on the telephone and microphone speech of the latest NIST 2010 SRE corpus show that the asymmetric-shaped tapers perform better than the symmetric Hamming window.

Full Paper

Bibliographic reference.  Alam, Md Jahangir / Kenny, Patrick / O'Shaughnessy, Douglas (2012): "On the use of asymmetric-shaped tapers for speaker verification using i-vectors", In Odyssey-2012, 256-262.