Maximization of the modelisation error ratio for neural predictive coding

In this paper, we introduce a model for Discrimant Feature Extraction (DFE): the Neural Predictive Coding (NPC). It is an extension of the Linear Predictive Coding (LPC). The Modelisation Error Ratio (MER), a discriminant criterion adapted for predictive models, is introduced. We propose a theoretical validation of the discriminant properties of the MER. The experimental validation consists on phoneme recognition task. The phonemes are extracted from the Darpa-Timit speech database. The performances are compared with traditional methods: LPC, MFCC, PLP.