Language model cross adaptation for LVCSR system combination

State-of-the-art large vocabulary continuous speech recognition (LVCSR) systems often combine outputs from multiple sub-systems that may even be developed at different sites. Cross system adaptation, in which model adaptation is performed using the outputs from another sub-system, can be used as an alternative to hypothesis level combination schemes such as ROVER. Normally cross adaptation is only performed on the acoustic models. However, there are many other levels in LVCSR systems' modelling hierarchy where complimentary features may be exploited, for example, the sub-word and the word level, to further improve cross adaptation based system combination. It is thus interesting to also cross adapt language models (LMs) to capture these additional useful features. In this paper cross adaptation is applied to three forms of language models, a multi-level LM that models both syllable and word sequences, a word level neural network LM, and the linear combination of the two. Significant error rate reductions of 4.0-7.1% relative were obtained over ROVER and acoustic model only cross adaptation when combining a range of Chinese LVCSR sub-systems used in the 2010 and 2011 DARPA GALE evaluations. © 2012 Elsevier Ltd. All rights reserved.

Shock transmission in a coupled beam system

This paper investigates the circumstances under which high peak acceleration can occur in the internal parts of a system when subjected to impulsive driving on the outside. Motivating examples include the design of packaging for transportation of fragile items. The system is modelled in an idealised form using two beams coupled with point connections. A Rayleigh-Ritz model of such coupled beams was validated against measurements on a particular beam system, then the model was used to explore the acceleration response to impulsive driving in the time, frequency and spatial domains. This study is restricted to linear vibration response and additional mechanisms for high internal acceleration due to nonlinear effects such as internal impacts are not considered. Using Monte Carlo simulation in which the indirectly driven beam was perturbed by randomly placed point masses a wide range of system behaviour was explored. This facilitates identification of vulnerable configurations that can lead to high internal acceleration. The results from the study indicate the possibility of curve veering influencing the peak acceleration amplification. The possibility of veering within an ensemble was found to be dependent on the relative coupling strength of the modes. Understanding of the mechanism may help to avoid vulnerable cases, either by design or by preparatory vibration testing. © 2013 Elsevier Ltd.

Sensorless commutation of a two-phase linear machine for marine renewable power generation

A sensorless scheme is presented for a two-phase permanent-magnet linear machine targeted for use in marine wave-power generation. This is a field where system reliability is a key concern. The scheme is able to extract the effective inductance and back-emf of the machine's phases simultaneously from measurements of the current ripple present on the power electronic converter. These measurements can then be used to estimate position. An enhancement to the scheme in the presence of spatially-varying mutual inductance between phases allows more accurate and reliable tracking from indutance-based measurements than would otherwise be expected. This scheme is able to operate at any speed including, critically, when stationary. Experimental results show promise for the scheme, although some work to reduce the level of noise would be desirable. © 2013 IEEE.

Encoding of mixtures in a simple olfactory system.

Natural odors are usually mixtures; yet, humans and animals can experience them as unitary percepts. Olfaction also enables stimulus categorization and generalization. We studied how these computations are performed with the responses of 168 locust antennal lobe projection neurons (PNs) to varying mixtures of two monomolecular odors, and of 174 PNs and 209 mushroom body Kenyon cells (KCs) to mixtures of up to eight monomolecular odors. Single-PN responses showed strong hypoadditivity and population trajectories clustered by odor concentration and mixture similarity. KC responses were much sparser on average than those of PNs and often signaled the presence of single components in mixtures. Linear classifiers could read out the responses of both populations in single time bins to perform odor identification, categorization, and generalization. Our results suggest that odor representations in the mushroom body may result from competing optimization constraints to facilitate memorization (sparseness) while enabling identification, classification, and generalization.

Encoding of mixtures in a simple olfactory system

Natural odors are usually mixtures; yet, humans and animals can experience them as unitary percepts. Olfaction also enables stimulus categorization and generalization. We studied how these computations are performed with the responses of 168 locust antennal lobe projection neurons (PNs) to varying mixtures of two monomolecular odors, and of 174 PNs and 209 mushroom body Kenyon cells (KCs) to mixtures of up to eight monomolecular odors. Single-PN responses showed strong hypoadditivity and population trajectories clustered by odor concentration and mixture similarity. KC responses were much sparser on average than those of PNs and often signaled the presence of single components in mixtures. Linear classifiers could read out the responses of both populations in single time bins to perform odor identification, categorization, and generalization. Our results suggest that odor representations in the mushroom body may result from competing optimization constraints to facilitate memorization (sparseness) while enabling identification, classification, and generalization

Kernelized log linear models for continuous speech recognition

Large margin criteria and discriminative models are two effective improvements for HMM-based speech recognition. This paper proposed a large margin trained log linear model with kernels for CSR. To avoid explicitly computing in the high dimensional feature space and to achieve the nonlinear decision boundaries, a kernel based training and decoding framework is proposed in this work. To make the system robust to noise a kernel adaptation scheme is also presented. Previous work in this area is extended in two directions. First, most kernels for CSR focus on measuring the similarity between two observation sequences. The proposed joint kernels defined a similarity between two observation-label sequence pairs on the sentence level. Second, this paper addresses how to efficiently employ kernels in large margin training and decoding with lattices. To the best of our knowledge, this is the first attempt at using large margin kernel-based log linear models for CSR. The model is evaluated on a noise corrupted continuous digit task: AURORA 2.0. © 2013 IEEE.

Field programmable gate array based predictive control system for spacecraft rendezvous in elliptical orbits

Copyright © 2014 John Wiley & Sons, Ltd. Copyright © 2014 John Wiley & Sons, Ltd. Summary A field programmable gate array (FPGA) based model predictive controller for two phases of spacecraft rendezvous is presented. Linear time-varying prediction models are used to accommodate elliptical orbits, and a variable prediction horizon is used to facilitate finite time completion of the longer range manoeuvres, whilst a fixed and receding prediction horizon is used for fine-grained tracking at close range. The resulting constrained optimisation problems are solved using a primal-dual interior point algorithm. The majority of the computational demand is in solving a system of simultaneous linear equations at each iteration of this algorithm. To accelerate these operations, a custom circuit is implemented, using a combination of Mathworks HDL Coder and Xilinx System Generator for DSP, and used as a peripheral to a MicroBlaze soft-core processor on the FPGA, on which the remainder of the system is implemented. Certain logic that can be hard-coded for fixed sized problems is implemented to be configurable online, in order to accommodate the varying problem sizes associated with the variable prediction horizon. The system is demonstrated in closed-loop by linking the FPGA with a simulation of the spacecraft dynamics running in Simulink on a PC, using Ethernet. Timing comparisons indicate that the custom implementation is substantially faster than pure embedded software-based interior point methods running on the same MicroBlaze and could be competitive with a pure custom hardware implementation.

Preloaded hopping with linear multi-modal actuation

For more dexterous and agile legged robot locomotion, alternative actuation has been one of the most long-awaited technologies. The goal of this paper is to investigate the use of newly developed actuator, the so-called Linear Multi-Modal Actuator (LMMA), in the context of legged robot locomotion, and analyze the behavioral performance of it. The LMMA consists of three discrete couplings which enable the system to switch between different mechanical dynamics such as instantaneous switches between series elastic and fully actuated dynamics. To test this actuator for legged locomotion, this paper introduces a one-legged robot platform we developed to implement the actuator, and explains a novel control strategy for hopping, i.e. 'preloaded hopping control'. This control strategy takes advantage of the coupling mechanism of the LMMA to preload the series elasticity during the flight phase to improve the energy efficiency of hopping locomotion. This paper shows a series of experimental results that compare the control strategy with a simple sinusoidal actuation strategy to discuss the benefits and challenges of the proposed approach. © 2013 IEEE.