Représentation acoustique compacte pour un système de reconnaissance de la parole embarquée
Christophe Lévy, Georges Linarès, Jean-François Bonastre.
Speech recognition applications are known to require a significant amount of resources (training data, memory, computing power). However, the targeted context of this work -mobile phone embedded speech recognition system- only authorizes few resources. In order to fit the resource constraints, an approach based on a HMM system using a GMM-based state-independent acoustic modeling is proposed in this paper. A transformation is computed and applied to the global GMM in order to obtain each of the HMM state-dependent probability density functions.
The proposed approach is evaluated on a French digit recognition task. Our method leads a Digit Error Rate (DER) of 2%, when the system respects the resource constraints. Compared to an HMM with comparable resource, our approach achieved a DER relative decrease more than 50%.