Constrained discriminative mapping transforms for unsupervised speaker adaptation

Discriminative mapping transforms (DMTs) is an approach to robustly adding discriminative training to unsupervised linear adaptation transforms. In unsupervised adaptation DMTs are more robust to unreliable transcriptions than directly estimating adaptation transforms in a discriminative fashion. They were previously proposed for use with MLLR transforms with the associated need to explicitly transform the model parameters. In this work the DMT is extended to CMLLR transforms. As these operate in the feature space, it is only necessary to apply a different linear transform at the front-end rather than modifying the model parameters. This is useful for rapidly changing speakers/environments. The performance of DMTs with CMLLR was evaluated on the WSJ 20k task. Experimental results show that DMTs based on constrained linear transforms yield 3% to 6% relative gain over MLE transforms in unsupervised speaker adaptation. © 2011 IEEE.

A global shallow water model using high order multi-moment constrained finite volume method and icosahedral grid

A novel accurate numerical model for shallow water equations on sphere have been developed by implementing the high order multi-moment constrained finite volume (MCV) method on the icosahedral geodesic grid. High order reconstructions are conducted cell-wisely by making use of the point values as the unknowns distributed within each triangular cell element. The time evolution equations to update the unknowns are derived from a set of constrained conditions for two types of moments, i.e. the point values on the cell boundary edges and the cell-integrated average. The numerical conservation is rigorously guaranteed. in the present model, all unknowns or computational variables are point values and no numerical quadrature is involved, which particularly benefits the computational accuracy and efficiency in handling the spherical geometry, such as coordinate transformation and curved surface. Numerical formulations of third and fourth order accuracy are presented in detail. The proposed numerical model has been validated by widely used benchmark tests and competitive results are obtained. The present numerical framework provides a promising and practical base for further development of atmospheric and oceanic general circulation models. (C) 2009 Elsevier Inc. All rights reserved.

adaptive location update mechanism for network-constrained moving objects in changeful traffic conditions

Natl Chiao Tung Univ, Dept Comp Sci

Estimation of gas hydrate saturation using constrained sparse spike inversion: Case study from the northern South China Sea

Bottom-simulating reflectors (BSRs) were observed beneath the seafloor in the northern continental margin of the South China Sea (SCS). Acoustic impedance profile was derived by Constrained Sparse Spike Inversion (CSSI) method to provide information on rock properties and to estimate gas hydrate or free gas saturations in the sediments where BSRs are present. In general, gas hydrate-bearing sediments have positive impedance anomalies and free gas-bearing sediments have negative impedance anomalies. Based on well log data and Archie's equation, gas hydrate saturation can be estimated. But in regions where well log data is not available, a quantitative estimate of gas hydrate or free gas saturation is inferred by fitting the theoretical acoustic impedance to sediment impedance obtained by CSSI. Our study suggests that gas hydrate saturation in the Taixinan Basin is about 10 - 20% of the pore space, with the highest value of 50%, and free gas saturation below BSR is about 2 - 3% of the pore space, that can rise to 8 - 10% at a topographic high. The free gas is non-continuous and has low content in the southeastern slope of the Dongsha Islands. Moreover, BSR in the northern continental margin of the SCS is related to the presence of free gas. BSR is strong where free gas occurs.