A novel estimation of feature-space MLLR for full-covariance models

In this paper we present a novel approach for estimating featurespace maximum likelihood linear regression (fMLLR) transforms for full-covariance Gaussian models by directly maximizing the likelihood function by repeated line search in the direction of the gradient. We do this in a pre-transformed parameter space such that an approximation to the expected Hessian is proportional to the unit matrix. The proposed algorithm is as efficient or more efficient than standard approaches, and is more flexible because it can naturally be combined with sets of basis transforms and with full covariance and subspace precision and mean (SPAM) models.

• Type: Conference
• Year: 2010
• Where: ICASSP
• Authors: Arnab Ghoshal, Daniel Povey, Mohit Agarwal, Pinar Akyazi, Lukas Burget, Kai Feng, Ondrej Glembek, Nagendra Goel, Martin Karafiát, Ariya Rastrow, Richard C. Rose, Petr Schwarz, Samuel Thomas
• Keywords: Featurespace Maximum Likelihood, Full-covariance Gaussian Models, ICASSP 2010, Pre-transformed Parameter Space, Signal Processing