Consistent DNN Uncertainty Training and Decoding for Robust ASR - Archive ouverte HAL Accéder directement au contenu
Communication Dans Un Congrès Année : 2017

Consistent DNN Uncertainty Training and Decoding for Robust ASR

Résumé

We consider the problem of robust automatic speech recognition (ASR) in noisy conditions. The performance improvement brought by speech enhancement is often limited by residual distortions of the enhanced features, which can be seen as a form of statistical uncertainty. Uncertainty estimation and propagation methods have recently been proposed to improve the ASR performance with deep neural network (DNN) acoustic models. However, the performance is still limited due to the use of uncertainty only during decoding. In this paper, we propose a consistent approach to account for uncertainty in the enhanced features during both training and decoding. We estimate the variance of the distortions using a DNN uncertainty estimator that operates directly in the feature maximum likelihood linear regression (fMLLR) domain and we then sample the uncertain features using the unscented transform (UT). We report the resulting ASR performance on the CHiME-2 and CHiME-3 datasets for different uncertainty estimation/propagation techniques. The proposed DNN uncertainty training method brings 4% and 8% relative improvement on these two datasets, respectively, compared to a competitive fMLLR-domain DNN acoustic modeling baseline.
Fichier principal
Vignette du fichier
Camera_Ready_ASRU.pdf (421.01 Ko) Télécharger le fichier
Origine : Fichiers produits par l'(les) auteur(s)
Loading...

Dates et versions

hal-01585956 , version 1 (12-09-2017)

Identifiants

  • HAL Id : hal-01585956 , version 1

Citer

Karan Nathwani, Emmanuel Vincent, Irina Illina. Consistent DNN Uncertainty Training and Decoding for Robust ASR. 2017 IEEE Automatic Speech Recognition and Understanding Workshop (ASRU), Dec 2017, Okinawa, Japan. ⟨hal-01585956⟩
907 Consultations
333 Téléchargements

Partager

Gmail Facebook X LinkedIn More