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.

Networked robots : flying robot navigation using a sensor net

This paper introduces the application of a sensor network to navigate a flying robot. We have developed distributed algorithms and efficient geographic routing techniques to incrementally guide one or more robots to points of interest based on sensor gradient fields, or along paths defined in terms of Cartesian coordinates. The robot itself is an integral part of the localization process which establishes the positions of sensors which are not known a priori. We use this system in a large-scale outdoor experiment with Mote sensors to guide an autonomous helicopter along a path encoded in the network. A simple handheld device, using this same environmental infrastructure, is used to guide humans.

Clustered blind beamforming from ad-hoc microphone arrays

Microphone arrays have been used in various applications to capture conversations, such as in meetings and teleconferences. In many cases, the microphone and likely source locations are known \emph{a priori}, and calculating beamforming filters is therefore straightforward. In ad-hoc situations, however, when the microphones have not been systematically positioned, this information is not available and beamforming must be achieved blindly. In achieving this, a commonly neglected issue is whether it is optimal to use all of the available microphones, or only an advantageous subset of these. This paper commences by reviewing different approaches to blind beamforming, characterising them by the way they estimate the signal propagation vector and the spatial coherence of noise in the absence of prior knowledge of microphone and speaker locations. Following this, a novel clustered approach to blind beamforming is motivated and developed. Without using any prior geometrical information, microphones are first grouped into localised clusters, which are then ranked according to their relative distance from a speaker. Beamforming is then performed using either the closest microphone cluster, or a weighted combination of clusters. The clustered algorithms are compared to the full set of microphones in experiments on a database recorded on different ad-hoc array geometries. These experiments evaluate the methods in terms of signal enhancement as well as performance on a large vocabulary speech recognition task.

The application of phonetic distribution normalisation to likelihood-maximising speech enhancement for robust ASR

Traditional speech enhancement methods optimise signal-level criteria such as signal-to-noise ratio, but such approaches are sub-optimal for noise-robust speech recognition. Likelihood-maximising (LIMA) frameworks on the other hand, optimise the parameters of speech enhancement algorithms based on state sequences generated by a speech recogniser for utterances of known transcriptions. Previous applications of LIMA frameworks have generated a set of global enhancement parameters for all model states without taking in account the distribution of model occurrence, making optimisation susceptible to favouring frequently occurring models, in particular silence. In this paper, we demonstrate the existence of highly disproportionate phonetic distributions on two corpora with distinct speech tasks, and propose to normalise the influence of each phone based on a priori occurrence probabilities. Likelihood analysis and speech recognition experiments verify this approach for improving ASR performance in noisy environments.

Unsupervised alignment of thousands of images

The task addressed in this thesis is the automatic alignment of an ensemble of misaligned images in an unsupervised manner. This application is especially useful in computer vision applications where annotations of the shape of an object of interest present in a collection of images is required. Performing this task manually is a slow, tedious, expensive and error prone process which hinders the progress of research laboratories and businesses. Most recently, the unsupervised removal of geometric variation present in a collection of images has been referred to as congealing based on the seminal work of Learned-Miller [21]. The only assumption made in congealing is that the parametric nature of the misalignment is known a priori (e.g. translation, similarity, a�ne, etc) and that the object of interest is guaranteed to be present in each image. The capability to congeal an ensemble of misaligned images stemming from the same object class has numerous applications in object recognition, detection and tracking. This thesis concerns itself with the construction of a congealing algorithm titled, least-squares congealing, which is inspired by the well known image to image alignment algorithm developed by Lucas and Kanade [24]. The algorithm is shown to have superior performance characteristics when compared to previously established methods: canonical congealing by Learned-Miller [21] and stochastic congealing by Z�ollei [39].

Experimental evaluation of hotspot identification methods

Identifying crash “hotspots”, “blackspots”, “sites with promise”, or “high risk” locations is standard practice in departments of transportation throughout the US. The literature is replete with the development and discussion of statistical methods for hotspot identification (HSID). Theoretical derivations and empirical studies have been used to weigh the benefits of various HSID methods; however, a small number of studies have used controlled experiments to systematically assess various methods. Using experimentally derived simulated data—which are argued to be superior to empirical data, three hot spot identification methods observed in practice are evaluated: simple ranking, confidence interval, and Empirical Bayes. Using simulated data, sites with promise are known a priori, in contrast to empirical data where high risk sites are not known for certain. To conduct the evaluation, properties of observed crash data are used to generate simulated crash frequency distributions at hypothetical sites. A variety of factors is manipulated to simulate a host of ‘real world’ conditions. Various levels of confidence are explored, and false positives (identifying a safe site as high risk) and false negatives (identifying a high risk site as safe) are compared across methods. Finally, the effects of crash history duration in the three HSID approaches are assessed. The results illustrate that the Empirical Bayes technique significantly outperforms ranking and confidence interval techniques (with certain caveats). As found by others, false positives and negatives are inversely related. Three years of crash history appears, in general, to provide an appropriate crash history duration.

Modeling crash types: New insights into the effects of covariates on crashes at rural intersections

Many studies focused on the development of crash prediction models have resulted in aggregate crash prediction models to quantify the safety effects of geometric, traffic, and environmental factors on the expected number of total, fatal, injury, and/or property damage crashes at specific locations. Crash prediction models focused on predicting different crash types, however, have rarely been developed. Crash type models are useful for at least three reasons. The first is motivated by the need to identify sites that are high risk with respect to specific crash types but that may not be revealed through crash totals. Second, countermeasures are likely to affect only a subset of all crashes—usually called target crashes—and so examination of crash types will lead to improved ability to identify effective countermeasures. Finally, there is a priori reason to believe that different crash types (e.g., rear-end, angle, etc.) are associated with road geometry, the environment, and traffic variables in different ways and as a result justify the estimation of individual predictive models. The objectives of this paper are to (1) demonstrate that different crash types are associated to predictor variables in different ways (as theorized) and (2) show that estimation of crash type models may lead to greater insights regarding crash occurrence and countermeasure effectiveness. This paper first describes the estimation results of crash prediction models for angle, head-on, rear-end, sideswipe (same direction and opposite direction), and pedestrian-involved crash types. Serving as a basis for comparison, a crash prediction model is estimated for total crashes. Based on 837 motor vehicle crashes collected on two-lane rural intersections in the state of Georgia, six prediction models are estimated resulting in two Poisson (P) models and four NB (NB) models. The analysis reveals that factors such as the annual average daily traffic, the presence of turning lanes, and the number of driveways have a positive association with each type of crash, whereas median widths and the presence of lighting are negatively associated. For the best fitting models covariates are related to crash types in different ways, suggesting that crash types are associated with different precrash conditions and that modeling total crash frequency may not be helpful for identifying specific countermeasures.

Microphone array beamforming approach to blind speech separation

In this paper, we present a microphone array beamforming approach to blind speech separation. Unlike previous beamforming approaches, our system does not require a-priori knowledge of the microphone placement and speaker location, making the system directly comparable other blind source separation methods which require no prior knowledge of recording conditions. Microphone location is automatically estimated using an assumed noise field model, and speaker locations are estimated using cross correlation based methods. The system is evaluated on the data provided for the PASCAL Speech Separation Challenge 2 (SSC2), achieving a word error rate of 58% on the evaluation set.

Governmentality and the reflection of legal educators : assessment practices as a case study

This paper presents a conceptual framework, informed by Foucault’s work on governmentality, which allows for new kinds of reflection on the practice of legal education. Put simply, this framework suggests that legal education can be understood as a form of government that relies on a specific rationalisation and programming of the activities of legal educators, students, and administrators, and is implemented by harnessing specific techniques and bodies of ‘know-how’. Applying this framework to assessment at three Australian law schools, this paper highlights how assessment practices are rationalised, programmed, and implemented, and points out how this government shapes students’ legal personae. In particular, this analysis focuses on the governmental effects of pedagogical discourses that are dominant within the design and scholarship of legal education. It demonstrates that the development of pedagogically-sound regimes of assessment has contributed to a reformulation of the terrain of government, by providing the conditions under which forms of legal personae may be more effectively shaped, and extending the power relations that achieve this. This analysis provides legal educators with an original way of reflecting on the power effects of teaching the law, and new opportunities for thinking about what is possible in legal education.

The Early Prevention of Obesity in CHildren (EPOCH) collaboration : an individual patient data prospective meta-analysis

Background: Efforts to prevent the development of overweight and obesity have increasingly focused early in the life course as we recognise that both metabolic and behavioural patterns are often established within the first few years of life. Randomised controlled trials (RCTs) of interventions are even more powerful when, with forethought, they are synthesised into an individual patient data (IPD) prospective meta-analysis (PMA). An IPD PMA is a unique research design where several trials are identified for inclusion in an analysis before any of the individual trial results become known and the data are provided for each randomised patient. This methodology minimises the publication and selection bias often associated with a retrospective meta-analysis by allowing hypotheses, analysis methods and selection criteria to be specified a priori. Methods/Design: The Early Prevention of Obesity in CHildren (EPOCH) Collaboration was formed in 2009. The main objective of the EPOCH Collaboration is to determine if early intervention for childhood obesity impacts on body mass index (BMI) z scores at age 18-24 months. Additional research questions will focus on whether early intervention has an impact on children’s dietary quality, TV viewing time, duration of breastfeeding and parenting styles. This protocol includes the hypotheses, inclusion criteria and outcome measures to be used in the IPD PMA. The sample size of the combined dataset at final outcome assessment (approximately 1800 infants) will allow greater precision when exploring differences in the effect of early intervention with respect to pre-specified participant- and intervention-level characteristics. Discussion: Finalisation of the data collection procedures and analysis plans will be complete by the end of 2010. Data collection and analysis will occur during 2011-2012 and results should be available by 2013. Trial registration number: ACTRN12610000789066

Creative industries after the first decade of debate

It has now been over a decade since the concept of creative industries was first put into the public domain through the Creative Industries Mapping Documents developed by the Blair Labour government in Britain. The concept has developed traction globally, but it has also been understood and developed in different ways in Europe, Asia, Australia, New Zealand and North America, as well as through international bodies such as UNCTAD and UNESCO. A review of the policy literature reveals that while questions and issues remain around definitional coherence, there is some degree of consensus emerging about the size, scope and significance of the sectors in question in both advanced and developing economies. At the same time, debate about the concept remains highly animated in media, communication and cultural studies, with its critics dismissing the concept outright as a harbinger of neo-liberal ideology in the cultural sphere. This paper couches such critiques in light of recent debates surrounding the intellectual coherence of the concept of neo-liberalism, arguing that this term itself possesses problems when taken outside of the Anglo-American context in which it originated. It is argued that issues surrounding the nature of participatory media culture, the relationship between cultural production and economic innovation, and the future role of public cultural institutions can be developed from within a creative industries framework, and that writing off such arguments as a priori ideological and flawed does little to advance debates about 21st century information and media culture.

Mapping the building and construction product system : preliminary report

This Preliminary Report has been prepared by researchers at The Australian Expert Group in Industry Studies (AEGIS) for the Commonwealth Department of Industry, Science and Resources. It is intended to provide a preliminary 'product system map' of the building and construction industries which defines the system, identifies the major segments, describes key industry players and institutions and provides the basis for exploring relationships, innovation and information flows within the industries. This Preliminary Report is the first of a series of five which will explore the building and construction product system in some depth. This first report does not present original research, although it does include some new interview data and analysis of a variety of written sources. This report is rather a reformulation of existing statistical and analytical material from a product system-based perspective. It is intended to provide the basis for subsequent studies by putting what is already known into an alternative framework and allowing us to see it through a new lens.

Queering Paradigms II : Interrogating Agendas

This book offers a fundamental challenge to a variety of theoretical, social, and political paradigms, ranging from law and justice studies to popular culture, linguistics to political activism. Developing the intellectual project initiated in Queering Paradigms, this volume extends queer theorizing in challenging new directions and uses queer insights to explore, trouble, and interrogate the social, political, and intellectual agendas that pervade (and are often taken for granted within) public discourses and academic disciplines. The contributing authors include queer theorists, socio-linguists, sociologists, political activists, educators, social workers and criminologists. Together, they contribute not only to the ongoing process of theorizing queerly, but also to the critique and reformulation of their respective disciplines.

Adaptive online gradient descent

We study the rates of growth of the regret in online convex optimization. First, we show that a simple extension of the algorithm of Hazan et al eliminates the need for a priori knowledge of the lower bound on the second derivatives of the observed functions. We then provide an algorithm, Adaptive Online Gradient Descent, which interpolates between the results of Zinkevich for linear functions and of Hazan et al for strongly convex functions, achieving intermediate rates between [square root T] and [log T]. Furthermore, we show strong optimality of the algorithm. Finally, we provide an extension of our results to general norms.

Barriers to information technology governance adoption : a preliminary empirical investigation

The adoption of IT Governance (ITG) continues to be an important topic for research. Many researchers have focused their attention on how these practices are currently being implemented in the many diverse areas and industries. Literature shows that a majority of these studies have only been based on industries and organizations in developed countries. There exist very few researches that look specifically within the context of a developing country. Furthermore, there seems to be a lack of research on identifying the barriers or inhibitors to IT Governance adoption within the context of an emerging yet still developing Asian country. This research sets out to justify, substantiate and improve on a priori model developed to study the barriers to the adoption of ITG practice using qualitative data obtained through a series of semi-structured interviews conducted on organizations in Malaysia.