Técnicas robustas para la discriminación de locutores

Abstract

Recently, a new filtering technique based on the decorrelation of filter bank energies has shown to be attractive for speech recognition because of its simplicity and its lower computational cost than standard representations like LPC-cepstrum or mel-cepstrum. Experimental results obtained in speaker identification show that this filtering technique is useful for discriminating speakers. On the other hand, different normalization techniques are applied and compared in speaker verification, and experimental results are shown.