Robust speech recognition using speech enhancement

Automatic Speech Recognition (ASR) has matured into a technology which is becoming more common in our everyday lives, and is emerging as a necessity to minimise driver distraction when operating in-car systems such as navigation and infotainment. In “noise-free” environments, word recognition performance of these systems has been shown to approach 100%, however this performance degrades rapidly as the level of background noise is increased. Speech enhancement is a popular method for making ASR systems more ro- bust. Single-channel spectral subtraction was originally designed to improve hu- man speech intelligibility and many attempts have been made to optimise this algorithm in terms of signal-based metrics such as maximised Signal-to-Noise Ratio (SNR) or minimised speech distortion. Such metrics are used to assess en- hancement performance for intelligibility not speech recognition, therefore mak- ing them sub-optimal ASR applications. This research investigates two methods for closely coupling subtractive-type enhancement algorithms with ASR: (a) a computationally-efficient Mel-filterbank noise subtraction technique based on likelihood-maximisation (LIMA), and (b) in- troducing phase spectrum information to enable spectral subtraction in the com- plex frequency domain. Likelihood-maximisation uses gradient-descent to optimise parameters of the enhancement algorithm to best fit the acoustic speech model given a word se- quence known a priori. Whilst this technique is shown to improve the ASR word accuracy performance, it is also identified to be particularly sensitive to non-noise mismatches between the training and testing data. Phase information has long been ignored in spectral subtraction as it is deemed to have little effect on human intelligibility. In this work it is shown that phase information is important in obtaining highly accurate estimates of clean speech magnitudes which are typically used in ASR feature extraction. Phase Estimation via Delay Projection is proposed based on the stationarity of sinusoidal signals, and demonstrates the potential to produce improvements in ASR word accuracy in a wide range of SNR. Throughout the dissertation, consideration is given to practical implemen- tation in vehicular environments which resulted in two novel contributions – a LIMA framework which takes advantage of the grounding procedure common to speech dialogue systems, and a resource-saving formulation of frequency-domain spectral subtraction for realisation in field-programmable gate array hardware. The techniques proposed in this dissertation were evaluated using the Aus- tralian English In-Car Speech Corpus which was collected as part of this work. This database is the first of its kind within Australia and captures real in-car speech of 50 native Australian speakers in seven driving conditions common to Australian environments.

Examination of social demographic status, disease status, health behaviours and quality of life in older Australians using the theory of planned behaviour model

The increase of life expectancy worldwide during the last three decades has increased age-related disability leading to the risk of loss of quality of life. How to improve quality of life including physical health and mental health for older people and optimize their life potential has become an important health issue. This study used the Theory of Planned Behaviour Model to examine factors influencing health behaviours, and the relationship with quality of life. A cross-sectional mailed survey of 1300 Australians over 50 years was conducted at the beginning of 2009, with 730 completed questionnaires returned (response rate 63%). Preliminary analysis reveals that physiological changes of old age, especially increasing waist circumference and co morbidity was closely related to health status, especially worse physical health summary score. Physical activity was the least adherent behaviour among the respondents compared to eating healthy food and taking medication regularly as prescribed. Increasing number of older people living alone with co morbidity of disease may be the barriers that influence their attitude and self control toward physical activity. A multidisciplinary and integrated approach including hospital and non hospital care is required to provide appropriate services and facilities toward older people.

Enhancing transition : an enhanced model of clinical placement for final year nursing students

Specialised support for student nurses making the transition to graduate nurse can be crucial to successful and smooth adjustment, and can create a path to positive and stable career experiences. This paper describes an enhanced model of final year nursing student placements which was trialled in 2006 at the Queensland University of Technology. The model involved collaboration with two major urban health services and resources were developed to support effective transition experiences. Ninety-two students, including 29 trial participants and 63 non-trial participants were assessed on preparedness for professional practice, before and after the trial semester. Results indicated an increase in preparedness across the entire sample, but students participating in the trial did not differ significantly in overall preparedness change from those who did not participate. Higher baseline preparedness in the trial group highlighted the possibility that proactive students who choose enrichment experiences tend to be likelier to gain benefit from such options than those who do not. Qualitative findings from focus groups conducted with 12 transition group students highlighted that one of the main beneficial aspects of the experience for students was the sense of belonging to a team that understood their learning needs and could work constructively with them.

Comparison of multiple and novel measures of dietary glycemic carbohydrate with insulin resistant status in older women

BACKGROUND:Previous epidemiological investigations of associations between dietary glycemic intake and insulin resistance have used average daily measures of glycemic index (GI) and glycemic load (GL). We explored multiple and novel measures of dietary glycemic intake to determine which was most predictive of an association with insulin resistance.METHODS:Usual dietary intakes were assessed by diet history interview in women aged 42-81 years participating in the Longitudinal Assessment of Ageing in Women. Daily measures of dietary glycemic intake (n = 329) were carbohydrate, GI, GL, and GL per megacalorie (GL/Mcal), while meal based measures (n = 200) were breakfast, lunch and dinner GL; and a new measure, GL peak score, to represent meal peaks. Insulin resistant status was defined as a homeostasis model assessment (HOMA) value of >3.99; HOMA as a continuous variable was also investigated.RESULTS:GL, GL/Mcal, carbohydrate (all P < 0.01), GL peak score (P = 0.04) and lunch GL (P = 0.04) were positively and independently associated with insulin resistant status. Daily measures were more predictive than meal-based measures, with minimal difference between GL/Mcal, GL and carbohydrate. No significant associations were observed with HOMA as a continuous variable.CONCLUSION:A dietary pattern with high peaks of GL above the individual's average intake was a significant independent predictor of insulin resistance in this population, however the contribution was less than daily GL and carbohydrate variables. Accounting for energy intake slightly increased the predictive ability of GL, which is potentially important when examining disease risk in more diverse populations with wider variations in energy requirements.

Improving automatic speaker verification using SVM techniques

Automatic recognition of people is an active field of research with important forensic and security applications. In these applications, it is not always possible for the subject to be in close proximity to the system. Voice represents a human behavioural trait which can be used to recognise people in such situations. Automatic Speaker Verification (ASV) is the process of verifying a persons identity through the analysis of their speech and enables recognition of a subject at a distance over a telephone channel { wired or wireless. A significant amount of research has focussed on the application of Gaussian mixture model (GMM) techniques to speaker verification systems providing state-of-the-art performance. GMM's are a type of generative classifier trained to model the probability distribution of the features used to represent a speaker. Recently introduced to the field of ASV research is the support vector machine (SVM). An SVM is a discriminative classifier requiring examples from both positive and negative classes to train a speaker model. The SVM is based on margin maximisation whereby a hyperplane attempts to separate classes in a high dimensional space. SVMs applied to the task of speaker verification have shown high potential, particularly when used to complement current GMM-based techniques in hybrid systems. This work aims to improve the performance of ASV systems using novel and innovative SVM-based techniques. Research was divided into three main themes: session variability compensation for SVMs; unsupervised model adaptation; and impostor dataset selection. The first theme investigated the differences between the GMM and SVM domains for the modelling of session variability | an aspect crucial for robust speaker verification. Techniques developed to improve the robustness of GMMbased classification were shown to bring about similar benefits to discriminative SVM classification through their integration in the hybrid GMM mean supervector SVM classifier. Further, the domains for the modelling of session variation were contrasted to find a number of common factors, however, the SVM-domain consistently provided marginally better session variation compensation. Minimal complementary information was found between the techniques due to the similarities in how they achieved their objectives. The second theme saw the proposal of a novel model for the purpose of session variation compensation in ASV systems. Continuous progressive model adaptation attempts to improve speaker models by retraining them after exploiting all encountered test utterances during normal use of the system. The introduction of the weight-based factor analysis model provided significant performance improvements of over 60% in an unsupervised scenario. SVM-based classification was then integrated into the progressive system providing further benefits in performance over the GMM counterpart. Analysis demonstrated that SVMs also hold several beneficial characteristics to the task of unsupervised model adaptation prompting further research in the area. In pursuing the final theme, an innovative background dataset selection technique was developed. This technique selects the most appropriate subset of examples from a large and diverse set of candidate impostor observations for use as the SVM background by exploiting the SVM training process. This selection was performed on a per-observation basis so as to overcome the shortcoming of the traditional heuristic-based approach to dataset selection. Results demonstrate the approach to provide performance improvements over both the use of the complete candidate dataset and the best heuristically-selected dataset whilst being only a fraction of the size. The refined dataset was also shown to generalise well to unseen corpora and be highly applicable to the selection of impostor cohorts required in alternate techniques for speaker verification.

Elucidating the links between UV radiation and vitamin D synthesis : using an in vitro model

Ultraviolet radiation (UV) is the carcinogen that causes the most common malignancy in humans – skin cancer. However, moderate UV exposure is essential for producing vitaminDin our skin. VitaminDincreases the absorption of calcium from the diet, and adequate calcium is necessary for the building and maintenance of bones. Thus, low levels of vitamin D can cause osteomalacia and rickets and contribute to osteoporosis. Emerging evidence also suggests vitamin D may protect against falls, internal cancers, psychiatric conditions, autoimmune diseases and cardiovascular diseases. Since the dominant source of vitamin D is sunlight exposure, there is a need to understand what is a “balanced” level of sun exposure to maintain an adequate level of vitamin D but minimise the risks of eye damage, skin damage and skin cancer resulting from excessive UV exposure. There are many steps in the pathway from incoming solar UV to the eventual vitamin D status of humans (measured as 25-hydroxyvitamin D in the blood), and our knowledge about many of these steps is currently incomplete. This project begins by investigating the levels of UV available for synthesising vitamin D, and how these levels vary across seasons, latitudes and times of the day. The thesis then covers experiments conducted with an in vitro model, which was developed to study several aspects of vitamin D synthesis. Results from the model suggest the relationship between UV dose and vitamin D is not linear. This is an important input into public health messages regarding ‘safe’ UV exposure: larger doses of UV, beyond a certain limit, may not continue to produce vitamin D; however, they will increase the risk of skin cancers and eye damage. The model also showed that, when given identical doses of UV, the amount of vitamin D produced was impacted by temperature. In humans, a temperature-dependent reaction must occur in the top layers of human skin, prior to vitamin D entering the bloodstream. The hypothesis will be raised that cooler temperatures (occurring in winter and at high latitudes) may reduce vitamin D production in humans. Finally, the model has also been used to study the wavelengths of UV thought to be responsible for producing vitamin D. It appears that vitamin D production is limited to a small range of UV wavelengths, which may be narrower than previously thought. Together, these results suggest that further research is needed into the ability of humans to synthesise vitamin D from sunlight. In particular, more information is needed about the dose-response relationship in humans and to investigate the proposed impact of temperature. Having an accurate action spectrum will also be essential for measuring the available levels of vitamin D-effective UV. As this research continues, it will contribute to the scientific evidence-base needed for devising a public health message that will balance the risks of excessive UV exposure with maintaining adequate vitamin D.

Determinants of participation restriction among community dwelling stroke survivors : a path analysis

Background: Apart from promoting physical recovery and assisting in activities of daily living, a major challenge in stroke rehabilitation is to minimize psychosocial morbidity and to promote the reintegration of stroke survivors into their family and community. The identification of key factors influencing long-term outcome are essential in developing more effective rehabilitation measures for reducing stroke-related morbidity. The aim of this study was to test a theoretical model of predictors of participation restriction which included the direct and indirect effects between psychosocial outcomes, physical outcome, and socio-demographic variables at 12 months after stroke.--------- Methods: Data were collected from 188 stroke survivors at 12 months following their discharge from one of the two rehabilitation hospitals in Hong Kong. The settings included patients' homes and residential care facilities. Path analysis was used to test a hypothesized model of participation restriction at 12 months.---------- Results: The path coefficients show functional ability having the largest direct effect on participation restriction (β = 0.51). The results also show that more depressive symptoms (β = -0.27), low state self-esteem (β = 0.20), female gender (β = 0.13), older age (β = -0.11) and living in a residential care facility (β = -0.12) have a direct effect on participation restriction. The explanatory variables accounted for 71% of the variance in explaining participation restriction at 12 months.---------- Conclusion: Identification of stroke survivors at risk of high levels of participation restriction, depressive symptoms and low self-esteem will assist health professionals to devise appropriate rehabilitation interventions that target improving both physical and psychosocial functioning.

Stroke education : promising effects on the health beliefs of those at risk

Background: The effect of patient education on reducing stroke has had mixed effects, raising questions about how to achieve optimal benefit. Because past evaluations have typically lacked an appropriate theoretical base, the design of past research may have missed important effects. --------- Method: This study used a social cognitive framework to identify variables that might change in response to education. A mixed design was used to evaluate two approaches to an intervention, both of which included education. Fifty seniors completed a measure of stroke knowledge and beliefs twice: before and after an intervention that was either standard (educational brochure plus activities that were not about stroke) or enhanced (educational brochure plus activities designed to enhance beliefs about stroke). Outcome measures were health beliefs, intention to exercise to reduce stroke, and stroke knowledge. --------- Results: Selected beliefs changed significantly over time but not differentially across conditions. Beliefs that changed were (a) perceived susceptibility to stroke and (b) perceived benefit of exercise to reduce risk. Benefit beliefs, in particular, were strongly and positively associated with intention to exercise. -------- Conclusion: Findings suggest that basic approaches to patient education may influence health beliefs. More effective stroke prevention programs may result from continued consideration of the role of health beliefs in such programs.

A Marxian model of technology appropriation

This paper explores the philosophical roots of appropriation within Marx's theories and socio-cultural studies in an attempt to seek common ground among existing theories of technology appropriation in IS research. Drawing on appropriation perspectives from Adaptive Structuration Theory, the Model of Technology Appropriation and the Structurational Model of Technology for comparison, we aim to generate a Marxian model that provides a starting point toward a general causal model of technology appropriation. This paper opens a philosophical discussion on the phenomenon of appropriation in the IS community, directing attention to foundational concepts in the human-technology nexus using ideas conceived by Marx.

An Integrative framework of the factors affecting process model understanding : a learning perspective

Process models are used by information professionals to convey semantics about the business operations in a real world domain intended to be supported by an information system. The understandability of these models is vital to them actually being used. After all, what is not understood cannot be acted upon. Yet until now, understandability has primarily been defined as an intrinsic quality of the models themselves. Moreover, those studies that looked at understandability from a user perspective have mainly conceptualized users through rather arbitrary sets of variables. In this paper we advance an integrative framework to understand the role of the user in the process of understanding process models. Building on cognitive psychology, goal-setting theory and multimedia learning theory, we identify three stages of learning required to realize model understanding, these being Presage, Process, and Product. We define eight relevant user characteristics in the Presage stage of learning, three knowledge construction variables in the Process stage and three potential learning outcomes in the Product stage. To illustrate the benefits of the framework, we review existing process modeling work to identify where our framework can complement and extend existing studies.

Data-driven background dataset selection for SVM-based speaker verification

The recently proposed data-driven background dataset refinement technique provides a means of selecting an informative background for support vector machine (SVM)-based speaker verification systems. This paper investigates the characteristics of the impostor examples in such highly-informative background datasets. Data-driven dataset refinement individually evaluates the suitability of candidate impostor examples for the SVM background prior to selecting the highest-ranking examples as a refined background dataset. Further, the characteristics of the refined dataset were analysed to investigate the desired traits of an informative SVM background. The most informative examples of the refined dataset were found to consist of large amounts of active speech and distinctive language characteristics. The data-driven refinement technique was shown to filter the set of candidate impostor examples to produce a more disperse representation of the impostor population in the SVM kernel space, thereby reducing the number of redundant and less-informative examples in the background dataset. Furthermore, data-driven refinement was shown to provide performance gains when applied to the difficult task of refining a small candidate dataset that was mis-matched to the evaluation conditions.

Modeling multivariate autoregressive conditional heteroskedasticity with the double smooth transition conditional correlation GARCH Model

In this paper, we propose a multivariate GARCH model with a time-varying conditional correlation structure. The new double smooth transition conditional correlation (DSTCC) GARCH model extends the smooth transition conditional correlation (STCC) GARCH model of Silvennoinen and Teräsvirta (2005) by including another variable according to which the correlations change smoothly between states of constant correlations. A Lagrange multiplier test is derived to test the constancy of correlations against the DSTCC-GARCH model, and another one to test for another transition in the STCC-GARCH framework. In addition, other specification tests, with the aim of aiding the model building procedure, are considered. Analytical expressions for the test statistics and the required derivatives are provided. Applying the model to the stock and bond futures data, we discover that the correlation pattern between them has dramatically changed around the turn of the century. The model is also applied to a selection of world stock indices, and we find evidence for an increasing degree of integration in the capital markets.