A text-dependent voice recognition approach using the spectral distance

The paper is discussing a novel speech-dependent speaker recognition approach. First, some related works in the voice-recognition domain are described, such as the auto-regression scheme. Next, we propose a Mel-Cepstral based speech analysis for voice feature extraction. Then, a special spectral distance is provided, to be used in speech feature vector classification process. This eigenvalue based metric represents the main novelty element of our paper. A supervised minimum mean distance classifier is also proposed by us. A high recognition rate is achieved using our speaker recognition technique, as resulting from our experiments.