Continuous speech recognition for the TIMIT database using neural networks

Four types of neural networks which have previously been established for speech recognition and tested on a small, seven-speaker, 100-sentence database are applied to the TIMIT database. The networks are a recurrent network phoneme recognizer, a modified Kanerva model morph recognizer, a compositional representation phoneme-to-word recognizer, and a modified Kanerva model morph-to-word recognizer. The major result is for the recurrent net, giving a phoneme recognition accuracy of 57% from the si and sx sentences. The Kanerva morph recognizer achieves 66.2% accuracy for a small subset of the sa and sx sentences. The results for the word recognizers are incomplete.

MMI training for continuous phoneme recognition on the TIMIT database

This paper reports our experiences with a phoneme recognition system for the TIMIT database which uses multiple mixture continuous density monophone HMMs trained using MMI. A comprehensive set of results are presented comparing the ML and MMI training criteria for both diagonal and full covariance models. These results using simple monophone HMMs show clear performance gains achieved by MMI training, and are comparable to the best reported by others including those which use context-dependent models. In addition, the paper discusses a number of performance and implementation issues which are crucial to successful MMI training.

An investigation into the residual stresses in an aluminium 2024 test weld

This paper uses finite element (FE) analysis to examine the residual stresses generated during the TIG welding of aluminium aerospace alloys. It also looks at whether such an approach could be useful for evaluating the effectiveness of various residual stress control techniques. However, such simulations cannot be founded in a vacuum. They require accurate measurements to refine and validate them. The unique aspect of this work is that two powerful engineering techniques are combined: FE modelling and neutron diffraction. Weld trials were performed and the direct measurement of residual strain made using the ENGIN neutron diffraction strain scanning facility. The predicted results show an excellent agreement with experimental values. Finally this model is used to simulate a weld made using a "Low Stress No Distortion" (LSND) technique. Although the stress reduction predicted is only moderate, the study suggests the approach to be a quick and efficient means of optimising such techniques.

Friction and material transfer in micro-scale sliding contact between aluminium alloy and steel

A ball-on-flat reciprocating micro-tribometer has been used to measure the friction coefficient between aluminium alloy strip and a steel ball. A relatively small ball and correspondingly low contact load is used to give a contact width of the order of 100μm, closer to asperity contact widths than generally found for this type of test. The effects of load, initial strip surface roughness, lubricants and boundary additives are investigated. It is found that the friction coefficient is significantly reduced by the addition of a lubricant. Observations of the wear tracks and ball surface show that the material transfer from aluminium to the ball is reduced in the presence of the lubricant. The initial friction coefficient is further reduced by the addition of a boundary additive, but the friction coefficient after 8 cycles is unchanged. Copyright © 2004 by Springer Science+Business Media, Inc.

Modelling of waterfront cantilever sheet pile walls subjected to earthquake loading

The seismic performance of waterfront cantilever sheet pile retaining walls is of continuing interest to geotechnical engineers as these structures suffer severe damage and even complete failure during earthquakes. This is often precipitated by liquefaction of the surrounding soil, either in the backfill or in front of the wall. This paper presents results from a series of small-scale plane strain models that were tested on a 1-g shaking table and recorded using a high-speed, high-resolution digital camera. The technique of Particle Image Velocimetry (PIV) was applied in order to allow the failure mechanisms to be visualised. It is shown that using PIV analyses it is possible to obtain failure mechanisms for a cantilever wall in liquefiable soil. These failure mechanisms are compared with those obtained for a cantilever wall in dry soil, previously carried out at a similar scale. It was observed that seismic liquefaction causes significant displacement in much larger zones of soil near the retaining wall compared to an equivalent dry case. The failure mechanism for a cantilever wall with liquefiable backfill, but with a remediated zone designed not to liquefy, is also presented and compared to the unremediated case.

Continuous F0 in the source-excitation generation for HMM-based TTS: Do we need voiced/unvoiced classification?

Most HMM-based TTS systems use a hard voiced/unvoiced classification to produce a discontinuous F0 signal which is used for the generation of the source-excitation. When a mixed source excitation is used, this decision can be based on two different sources of information: the state-specific MSD-prior of the F0 models, and/or the frame-specific features generated by the aperiodicity model. This paper examines the meaning of these variables in the synthesis process, their interaction, and how they affect the perceived quality of the generated speech The results of several perceptual experiments show that when using mixed excitation, subjects consistently prefer samples with very few or no false unvoiced errors, whereas a reduction in the rate of false voiced errors does not produce any perceptual improvement. This suggests that rather than using any form of hard voiced/unvoiced classification, e.g., the MSD-prior, it is better for synthesis to use a continuous F0 signal and rely on the frame-level soft voiced/unvoiced decision of the aperiodicity model. © 2011 IEEE.