Derivative kernels for noise robust ASR

Recently there has been interest in combining generative and discriminative classifiers. In these classifiers features for the discriminative models are derived from the generative kernels. One advantage of using generative kernels is that systematic approaches exist to introduce complex dependencies into the feature-space. Furthermore, as the features are based on generative models standard model-based compensation and adaptation techniques can be applied to make discriminative models robust to noise and speaker conditions. This paper extends previous work in this framework in several directions. First, it introduces derivative kernels based on context-dependent generative models. Second, it describes how derivative kernels can be incorporated in structured discriminative models. Third, it addresses the issues associated with large number of classes and parameters when context-dependent models and high-dimensional feature-spaces of derivative kernels are used. The approach is evaluated on two noise-corrupted tasks: small vocabulary AURORA 2 and medium-to-large vocabulary AURORA 4 task. © 2011 IEEE.

Measurement of the elastic constants of nanometric films

Carbon coatings of thickness down to 2 nanometers are needed to increase the storage density in magnetic hard disks and reach the 100 Gbit/in2 target. Methods to measure the properties of these ultrathin hard films still have to be developed. We show that combining Surface Brillouin Scattering (SBS) andX-ray reflectivity measurements the elastic constants of such films are accessible. Tetrahedral amorphous carbofilms of thickness down to about 2 nm were deposited on Si by an S bend filtered cathodic vacuum arc, achieving a continuous coverage on large areas free of macroparticles. Film thickness and mass density are measured by X-ray reflectivity: densities above 3 g/cm3 are found, indicating a significant sp3 content. The dispersion relations of surface acoustic waves are measured by SBS. We show that for thicknesses above ∼4 nm these waves can be described by a continuum elastic model based on a single homogeneous equivalent film. The elastic constants can then be obtained by fitting the dispersion relations, computed for given film properties, to the measured dispersion relations. For thicknesses of 3 nm or less qualitative differences among films are well measurable, but quantitative results are less reliable. We have thus shown that we can grow and characterise nanometer size tetrahedral amorphous carbon film, which maintain their high density and peculiar mechanical properties down to around 4 nm thickness, satisfying the requirements set for the hard disk coating material.

Self-stabilization and behavioral diversity of embodied adaptive locomotion

Locomotion is of fundamental importance in understanding adaptive behavior. In this paper we present two case studies of robot locomotion that demonstrate how higher level of behavioral diversity can be achieved while observing the principle of cheap design. More precisely, it is shown that, by exploiting the dynamics of the system-environment interaction, very simple controllers can be designed which is essential to achieve rapid locomotion. Special consideration must be given to the choice of body materials. We conclude with some speculation about the importance of locomotion for understanding cognition. © Springer-Verlag Berlin Heidelberg 2004.