Bayes factor based speaker clustering for speaker diarization

This paper proposes the use of the Bayes Factor to replace the Bayesian Information Criterion (BIC) as a criterion for speaker clustering within a speaker diarization system. The BIC is one of the most popular decision criteria used in speaker diarization systems today. However, it will be shown in this paper that the BIC is only an approximation to the Bayes factor of marginal likelihoods of the data given each hypothesis. This paper uses the Bayes factor directly as a decision criterion for speaker clustering, thus removing the error introduced by the BIC approximation. Results obtained on the 2002 Rich Transcription (RT-02) Evaluation dataset show an improved clustering performance, leading to a 14.7% relative improvement in the overall Diarization Error Rate (DER) compared to the baseline system.

Simulated cataracts and their effect on speech intelligibility

Purpose: The classic study of Sumby and Pollack (1954, JASA, 26(2), 212-215) demonstrated that visual information aided speech intelligibility under noisy auditory conditions. Their work showed that visual information is especially useful under low signal-to-noise conditions where the auditory signal leaves greater margins for improvement. We investigated whether simulated cataracts interfered with the ability of participants to use visual cues to help disambiguate the auditory signal in the presence of auditory noise. Methods: Participants in the study were screened to ensure normal visual acuity (mean of 20/20) and normal hearing (auditory threshold ≤ 20 dB HL). Speech intelligibility was tested under an auditory only condition and two visual conditions: normal vision and simulated cataracts. The light scattering effects of cataracts were imitated using cataract-simulating filters. Participants wore blacked-out glasses in the auditory only condition and lens-free frames in the normal auditory-visual condition. Individual sentences were spoken by a live speaker in the presence of prerecorded four-person background babble set to a speech-to-noise ratio (SNR) of -16 dB. The SNR was determined in a preliminary experiment to support 50% correct identification of sentence under the auditory only conditions. The speaker was trained to match the rate, intensity and inflections of a prerecorded audio track of everyday speech sentences. The speaker was blind to the visual conditions of the participant to control for bias.Participants’ speech intelligibility was measured by comparing the accuracy of their written account of what they believed the speaker to have said to the actual spoken sentence. Results: Relative to the normal vision condition, speech intelligibility was significantly poorer when participants wore simulated catarcts. Conclusions: The results suggest that cataracts may interfere with the acquisition of visual cues to speech perception.

Identification of alcohol involvement in injury-related hospitalisations using routine data compared to medical record review

Objective: To quantify the extent to which alcohol related injuries are adequately identified in hospitalisation data using ICD-10-AM codes indicative of alcohol involvement. Method: A random sample of 4373 injury-related hospital separations from 1 July 2002 to 30 June 2004 were obtained from a stratified random sample of 50 hospitals across 4 states in Australia. From this sample, cases were identified as involving alcohol if they contained an ICD-10-AM diagnosis or external cause code referring to alcohol, or if the text description extracted from the medical records mentioned alcohol involvement. Results: Overall, identification of alcohol involvement using ICD codes detected 38% of the alcohol-related sample, whilst almost 94% of alcohol-related cases were identified through a search of the text extracted from the medical records. The resultant estimate of alcohol involvement in injury-related hospitalisations in this sample was 10%. Emergency department records were the most likely to identify whether the injury was alcohol-related with almost three-quarters of alcohol-related cases mentioning alcohol in the text abstracted from these records. Conclusions and Implications: The current best estimates of the frequency of hospital admissions where alcohol is involved prior to the injury underestimate the burden by around 62%. This is a substantial underestimate that has major implications for public policy, and highlights the need for further work on improving the quality and completeness of routine administrative data sources for identification of alcohol-related injuries.

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.

QUT speaker identity verification system for EVALITA 2009

This document outlines the system submitted by the Speech and Audio Research Laboratory at the Queensland University of Technology (QUT) for the Speaker Identity Verication: Application task of EVALITA 2009. This submission consisted of a score-level fusion of three component systems, a joint-factor GMM system and two SVM systems using GLDS and GMM supervector kernels. Development and evaluation results are presented, demonstrating the effectiveness of this fused system approach.

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.

Exploiting multiple feature sets in data-driven impostor dataset selection for speaker verification

This study assesses the recently proposed data-driven background dataset refinement technique for speaker verification using alternate SVM feature sets to the GMM supervector features for which it was originally designed. The performance improvements brought about in each trialled SVM configuration demonstrate the versatility of background dataset refinement. This work also extends on the originally proposed technique to exploit support vector coefficients as an impostor suitability metric in the data-driven selection process. Using support vector coefficients improved the performance of the refined datasets in the evaluation of unseen data. Further, attempts are made to exploit the differences in impostor example suitability measures from varying features spaces to provide added robustness.

A preliminary investigation of the role of the Phenylalynine: Tyrosine Ratio in children with early and continuously treated phenylketonuria : toward identification of “safe” levels

Children with early and continuously treated phenylketonuria (ECT-PKU) remain at risk of developing executive function (EF) deficits. There is some evidence that a high phenylalanine to tyrosine ratio (phe:tyr) is more strongly associated with impaired EF development than high phenylalanine alone. This study examined EF in a sample of 11 adolescents against concurrent and historical levels of phenylalanine, phe:tyr, and tyrosine. Lifetime measures of phe:tyr were more strongly associated with EF than phenylalanine-only measures. Children with a lifetime phe:tyr less than 6 demonstrated normal EF, whereas children who had a lifetime phe:tyr above 6, on average, demonstrated clinically impaired EF.

System identification and control of an aerobot drive system

Fast thrust changes are important for authoritive control of VTOL micro air vehicles. Fixed-pitch rotors that alter thrust by varying rotor speed require high-bandwidth control systems to provide adequate performace. We develop a feedback compensator for a brushless hobby motor driving a custom rotor suitable for UAVs. The system plant is identified using step excitation experiments. The aerodynamic operating conditions of these rotors are unusual and so experiments are performed to characterise expected load disturbances. The plant and load models lead to a proportional controller design capable of significantly decreasing rise-time and propagation of disturbances, subject to bus voltage constraints.

Identification and specification of relationships as the foundation for service bundling

Service bundling can be regarded as an option for service providers to strengthen their competitive advantages, cope with dynamic market conditions and heterogeneous consumer demand. Despite these positive effects, actual guidance for the identification of service bundles and the act of bundling itself can be regarded as a gap. Previous research has resulted in a conceptualization of a service bundling method relying on a structured service description in order to fill this gap. This method addresses the reasoning about the suitability of services to be part of a bundle based on analyzing existing relationships between services captured by a description language. This paper extends the aforementioned research by presenting an initial set of empirically derived relationships between services in existing bundles that can subsequently be utilized to identify potential new bundles. Additionally, a gap analysis points out to what extent prominent ontologies and service description languages accommodate for the identified relationships.

Service identification through value chain analysis and prioritization

In a resource constrained business world, strategic choices must be made on process improvement and service delivery. There are calls for more agile forms of enterprises and much effort is being directed at moving organizations from a complex landscape of disparate application systems to that of an integrated and flexible enterprise accessing complex systems landscapes through service oriented architecture (SOA). This paper describes the deconstruction of an enterprise into business services using value chain analysis as each element in the value chain can be rendered as a business service in the SOA. These business services are explicitly linked to the attainment of specific organizational strategies and their contribution to the attainment of strategy is assessed and recorded. This contribution is then used to provide a rank order of business service to strategy. This information facilitates executive decision making on which business service to develop into the SOA. The paper describes an application of this Critical Service Identification Methodology (CSIM) to a case study.

An identification of factors influencing police workplace motivation

Police work tasks are diverse and require the ability to take command, demonstrate leadership, make serious decisions and be self directed (Beck, 1999; Brunetto & Farr-Wharton, 2002; Howard, Donofrio & Boles, 2002). This work is usually performed in pairs or sometimes by an officer working alone. Operational police work is seldom performed under the watchful eyes of a supervisor and a great amount of reliance is placed on the high levels of motivation and professionalism of individual officers. Research has shown that highly motivated workers produce better outcomes (Whisenand & Rush, 1998; Herzberg, 2003). It is therefore important that Queensland police officers are highly motivated to provide a quality service to the Queensland community. This research aims to identify factors which motivate Queensland police to perform quality work. Researchers acknowledge that there is a lack of research and knowledge in regard to the factors which motivate police (Beck, 1999; Bragg, 1998; Howard, Donofrio & Boles, 2002; McHugh & Verner, 1998). The motivational factors were identified in regard to the demographic variables of; age, sex, rank, tenure and education. The model for this research is Herzberg’s two-factor theory of workplace motivation (1959). Herzberg found that there are two broad types of workplace motivational factors; those driven by a need to prevent loss or harm and those driven by a need to gain personal satisfaction or achievement. His study identified 16 basic sub-factors that operate in the workplace. The research utilised a questionnaire instrument based on the sub-factors identified by Herzberg (1959). The questionnaire format consists of an initial section which sought demographic information about the participant and is followed by 51 Likert scale questions. The instrument is an expanded version of an instrument previously used in doctoral studies to identify sources of police motivation (Holden, 1980; Chiou, 2004). The questionnaire was forwarded to approximately 960 police in the Brisbane, Metropolitan North Region. The data were analysed using Factor Analysis, MANOVAs, ANOVAs and multiple regression analysis to identify the key sources of police motivation and to determine the relationships between demographic variables such as: age, rank, educational level, tenure, generation cohort and motivational factors. A total of 484 officers responded to the questionnaire from the sample population of 960. Factor analysis revealed five broad Prime Motivational Factors that motivate police in their work. The Prime Motivational Factors are: Feeling Valued, Achievement, Workplace Relationships, the Work Itself and Pay and Conditions. The factor Feeling Valued highlighted the importance of positive supportive leaders in motivating officers. Many officers commented that supervisors who only provided negative feedback diminished their sense of feeling valued and were a key source of de-motivation. Officers also frequently commented that they were motivated by operational police work itself whilst demonstrating a strong sense of identity with their team and colleagues. The study showed a general need for acceptance by peers and an idealistic motivation to assist members of the community in need and protect victims of crime. Generational cohorts were not found to exert a significant influence on police motivation. The demographic variable with the single greatest influence on police motivation was tenure. Motivation levels were found to drop dramatically during the first two years of an officer’s service and generally not improve significantly until near retirement age. The findings of this research provide the foundation of a number of recommendations in regard to police retirement, training and work allocation that are aimed to improve police motivation levels. The five Prime Motivational Factor model developed in this study is recommended for use as a planning tool by police leaders to improve motivational and job-satisfaction components of police Service policies. The findings of this study also provide a better understanding of the current sources of police motivation. They are expected to have valuable application for Queensland police human resource management when considering policies and procedures in the areas of motivation, stress reduction and attracting suitable staff to specific areas of responsibility.

Alternative analytical methods for the identification of cancer-related symptom clusters

Advances in symptom management strategies through a better understanding of cancer symptom clusters depend on the identification of symptom clusters that are valid and reliable. The purpose of this exploratory research was to investigate alternative analytical approaches to identify symptom clusters for patients with cancer, using readily accessible statistical methods, and to justify which methods of identification may be appropriate for this context. Three studies were undertaken: (1) a systematic review of the literature, to identify analytical methods commonly used for symptom cluster identification for cancer patients; (2) a secondary data analysis to identify symptom clusters and compare alternative methods, as a guide to best practice approaches in cross-sectional studies; and (3) a secondary data analysis to investigate the stability of symptom clusters over time. The systematic literature review identified, in 10 years prior to March 2007, 13 cross-sectional studies implementing multivariate methods to identify cancer related symptom clusters. The methods commonly used to group symptoms were exploratory factor analysis, hierarchical cluster analysis and principal components analysis. Common factor analysis methods were recommended as the best practice cross-sectional methods for cancer symptom cluster identification. A comparison of alternative common factor analysis methods was conducted, in a secondary analysis of a sample of 219 ambulatory cancer patients with mixed diagnoses, assessed within one month of commencing chemotherapy treatment. Principal axis factoring, unweighted least squares and image factor analysis identified five consistent symptom clusters, based on patient self-reported distress ratings of 42 physical symptoms. Extraction of an additional cluster was necessary when using alpha factor analysis to determine clinically relevant symptom clusters. The recommended approaches for symptom cluster identification using nonmultivariate normal data were: principal axis factoring or unweighted least squares for factor extraction, followed by oblique rotation; and use of the scree plot and Minimum Average Partial procedure to determine the number of factors. In contrast to other studies which typically interpret pattern coefficients alone, in these studies symptom clusters were determined on the basis of structure coefficients. This approach was adopted for the stability of the results as structure coefficients are correlations between factors and symptoms unaffected by the correlations between factors. Symptoms could be associated with multiple clusters as a foundation for investigating potential interventions. The stability of these five symptom clusters was investigated in separate common factor analyses, 6 and 12 months after chemotherapy commenced. Five qualitatively consistent symptom clusters were identified over time (Musculoskeletal-discomforts/lethargy, Oral-discomforts, Gastrointestinaldiscomforts, Vasomotor-symptoms, Gastrointestinal-toxicities), but at 12 months two additional clusters were determined (Lethargy and Gastrointestinal/digestive symptoms). Future studies should include physical, psychological, and cognitive symptoms. Further investigation of the identified symptom clusters is required for validation, to examine causality, and potentially to suggest interventions for symptom management. Future studies should use longitudinal analyses to investigate change in symptom clusters, the influence of patient related factors, and the impact on outcomes (e.g., daily functioning) over time.