Hybrid KLT/GMM approach for robust speaker identification
C.C.T. Chen, C.T. Chen, Philip Wing Ping Cheng · Electronics Letters · 2003
A classification scheme that incorporates both the Karhunen–Loeve transform and the Gaussian mixture model for robust speaker identification is proposed. Two hundred speakers from the TIMIT database degraded by noise are used for system evaluation. It is demonstrated that accuracy improvement up to 41% and computational saving up to seven times compared to the conventional GMM model can be achieved.