ESCA Workshop on Automatic Speaker Recognition, Identification, and Verification Martigny, Switzerland April 7-9, 1994

Using Higher Order Statistics to Increase the Noise Robustness of a Speaker Identification System

Larry J. Trent (1), Charles M. Rader (2), Douglas A. Reynolds (2)

(1) Technology Applications Office at Hanscom Air Force Base, USA
(2) Massachusetts Institute of Technology Lincoln Laboratory, Lexington, MA, USA

At present, most speaker identification systems use cepstral or linear prediction (IP) based features. The performance of these systems degrades significantly with the presence of noise in the training and/or the testing speech. To improve this performance, higher order statistics (HOS) or cumulant-based LF features are proposed. Using these features and singular value decomposition (SVD), it is shown that the performance of a speaker identification system can be improved considerably in the presence of additive white or colored Gaussian noise.

Full Paper

Bibliographic reference.  Trent, Larry J. / Rader, Charles M. / Reynolds, Douglas A. (1994): "Using higher order statistics to increase the noise robustness of a speaker identification system", In ASRIV-1994, 221-224.