Uncertainty propagation through deep neural networks

Abstract : In order to improve the ASR performance in noisy environments , distorted speech is typically pre-processed by a speech enhancement algorithm, which usually results in a speech estimate containing residual noise and distortion. We may also have some measures of uncertainty or variance of the estimate. Uncertainty decoding is a framework that utilizes this knowledge of uncertainty in the input features during acoustic model scoring. Such frameworks have been well explored for traditional probabilistic models, but their optimal use for deep neural network (DNN)-based ASR systems is not yet clear. In this paper, we study the propagation of observation uncertainties through the layers of a DNN-based acoustic model. Since this is intractable due to the nonlinearities of the DNN, we employ approximate propagation methods, including Monte Carlo sampling , the unscented transform, and the piecewise exponential approximation of the activation function, to estimate the distribution of acoustic scores. Finally, the expected value of the acoustic score distribution is used for decoding, which is shown to further improve the ASR accuracy on the CHiME database, relative to a highly optimized DNN baseline.
Type de document :
Communication dans un congrès
Interspeech 2015, Sep 2015, Dresden, Germany. 2015
Liste complète des métadonnées

Contributeur : Emmanuel Vincent <>
Soumis le : mercredi 10 juin 2015 - 18:06:48
Dernière modification le : jeudi 11 janvier 2018 - 06:27:31
Document(s) archivé(s) le : mardi 25 avril 2017 - 06:31:56


Fichiers produits par l'(les) auteur(s)


  • HAL Id : hal-01162550, version 1


Ahmed H. Abdelaziz, Shinji Watanabe, John R. Hershey, Emmanuel Vincent, Dorothea Kolossa. Uncertainty propagation through deep neural networks. Interspeech 2015, Sep 2015, Dresden, Germany. 2015. 〈hal-01162550〉



Consultations de la notice


Téléchargements de fichiers