Skip to Main content Skip to Navigation
New interface
Reports (Research report)

Uncertainty-based learning of Gaussian mixture models from noisy data

Alexey Ozerov 1 Mathieu Lagrange 2 Emmanuel Vincent 3 
3 METISS - Speech and sound data modeling and processing
IRISA - Institut de Recherche en Informatique et Systèmes Aléatoires, Inria Rennes – Bretagne Atlantique
Abstract : We consider the problem of Gaussian mixture model (GMM)-based classification of noisy data, where the uncertainty over the data is given by a Gaussian distribution. While this uncertainty is commonly exploited at the decoding stage via uncertainty decoding, it has not been exploited at the training stage so far. We introduce a new Expectation-Maximization (EM) algorithm called uncertainty training that allows to learn GMMs directly from noisy data while taking their uncertainty into account. We evaluate its potential for a speaker recognition task over speech data corrupted by real-world domestic background noise, using a state-of-the-art signal enhancement technique and various uncertainty estimation techniques as a front-end. Compared to conventional training, the proposed algorithm results in 3\% to 4\% absolute improvement in speaker recognition accuracy by training from either matched, unmatched or multi-condition noisy data. This algorithm is also applicable with minor modifications to maximum a posteriori (MAP) or maximum likelihood linear regression (MLLR) model adaptation and to the training of hidden Markov models (HMMs) from noisy data.
Complete list of metadata
Contributor : Emmanuel Vincent Connect in order to contact the contributor
Submitted on : Sunday, July 15, 2012 - 3:16:40 PM
Last modification on : Thursday, October 27, 2022 - 3:45:17 AM
Long-term archiving on: : Tuesday, October 16, 2012 - 2:21:28 AM


Files produced by the author(s)


  • HAL Id : hal-00660689, version 2


Alexey Ozerov, Mathieu Lagrange, Emmanuel Vincent. Uncertainty-based learning of Gaussian mixture models from noisy data. [Research Report] RR-7862, 2012. ⟨hal-00660689v2⟩



Record views


Files downloads