System combination and score normalization for spoken term detection

Spoken content in languages of emerging importance needs to be searchable to provide access to the underlying information. In this paper, we investigate the problem of extending data fusion methodologies from Information Retrieval for Spoken Term Detection on low-resource languages in the framework of the IARPA Babel program. We describe a number of alternative methods improving keyword search performance. We apply these methods to Cantonese, a language that presents some new issues in terms of reduced resources and shorter query lengths. First, we show score normalization methodology that improves in average by 20% keyword search performance. Second, we show that properly combining the outputs of diverse ASR systems performs 14% better than the best normalized ASR system. © 2013 IEEE.

PLDA based speaker recognition on short utterances

This paper investigates the effects of limited speech data in the context of speaker verification using a probabilistic linear discriminant analysis (PLDA) approach. Being able to reduce the length of required speech data is important to the development of automatic speaker verification system in real world applications. When sufficient speech is available, previous research has shown that heavy-tailed PLDA (HTPLDA) modeling of speakers in the i-vector space provides state-of-the-art performance, however, the robustness of HTPLDA to the limited speech resources in development, enrolment and verification is an important issue that has not yet been investigated. In this paper, we analyze the speaker verification performance with regards to the duration of utterances used for both speaker evaluation (enrolment and verification) and score normalization and PLDA modeling during development. Two different approaches to total-variability representation are analyzed within the PLDA approach to show improved performance in short-utterance mismatched evaluation conditions and conditions for which insufficient speech resources are available for adequate system development. The results presented within this paper using the NIST 2008 Speaker Recognition Evaluation dataset suggest that the HTPLDA system can continue to achieve better performance than Gaussian PLDA (GPLDA) as evaluation utterance lengths are decreased. We also highlight the importance of matching durations for score normalization and PLDA modeling to the expected evaluation conditions. Finally, we found that a pooled total-variability approach to PLDA modeling can achieve better performance than the traditional concatenated total-variability approach for short utterances in mismatched evaluation conditions and conditions for which insufficient speech resources are available for adequate system development.

Robust face clustering for real-world data

This thesis has investigated how to cluster a large number of faces within a multi-media corpus in the presence of large session variation. Quality metrics are used to select the best faces to represent a sequence of faces; and session variation modelling improves clustering performance in the presence of wide variations across videos. Findings from this thesis contribute to improving the performance of both face verification systems and the fully automated clustering of faces from a large video corpus.

Advanced Techniques for Face Recognition under Challenging Environments

Automatically recognizing faces captured under uncontrolled environments has always been a challenging topic in the past decades. In this work, we investigate cohort score normalization that has been widely used in biometric verification as means to improve the robustness of face recognition under challenging environments. In particular, we introduce cohort score normalization into undersampled face recognition problem. Further, we develop an effective cohort normalization method specifically for the unconstrained face pair matching problem. Extensive experiments conducted on several well known face databases demonstrate the effectiveness of cohort normalization on these challenging scenarios. In addition, to give a proper understanding of cohort behavior, we study the impact of the number and quality of cohort samples on the normalization performance. The experimental results show that bigger cohort set size gives more stable and often better results to a point before the performance saturates. And cohort samples with different quality indeed produce different cohort normalization performance. Recognizing faces gone after alterations is another challenging problem for current face recognition algorithms. Face image alterations can be roughly classified into two categories: unintentional (e.g., geometrics transformations introduced by the acquisition devide) and intentional alterations (e.g., plastic surgery). We study the impact of these alterations on face recognition accuracy. Our results show that state-of-the-art algorithms are able to overcome limited digital alterations but are sensitive to more relevant modifications. Further, we develop two useful descriptors for detecting those alterations which can significantly affect the recognition performance. In the end, we propose to use the Structural Similarity (SSIM) quality map to detect and model variations due to plastic surgeries. Extensive experiments conducted on a plastic surgery face database demonstrate the potential of SSIM map for matching face images after surgeries.

Hierarchical clustering of flow cytometry data for the study of conventional central chondrosarcoma

We have investigated the use of hierarchical clustering of flow cytometry data to classify samples of conventional central chondrosarcoma, a malignant cartilage forming tumor of uncertain cellular origin, according to similarities with surface marker profiles of several known cell types. Human primary chondrosarcoma cells, articular chondrocytes, mesenchymal stem cells, fibroblasts, and a panel of tumor cell lines from chondrocytic or epithelial origin were clustered based on the expression profile of eleven surface markers. For clustering, eight hierarchical clustering algorithms, three distance metrics, as well as several approaches for data preprocessing, including multivariate outlier detection, logarithmic transformation, and z-score normalization, were systematically evaluated. By selecting clustering approaches shown to give reproducible results for cluster recovery of known cell types, primary conventional central chondrosacoma cells could be grouped in two main clusters with distinctive marker expression signatures: one group clustering together with mesenchymal stem cells (CD49b-high/CD10-low/CD221-high) and a second group clustering close to fibroblasts (CD49b-low/CD10-high/CD221-low). Hierarchical clustering also revealed substantial differences between primary conventional central chondrosarcoma cells and established chondrosarcoma cell lines, with the latter not only segregating apart from primary tumor cells and normal tissue cells, but clustering together with cell lines from epithelial lineage. Our study provides a foundation for the use of hierarchical clustering applied to flow cytometry data as a powerful tool to classify samples according to marker expression patterns, which could lead to uncover new cancer subtypes.

Normalization procedure for the Baptista Depression Scale - Adult Version (EBADEP-A): transferring of norms

Regarding the standardization of psychological assessment instruments, that is, the construction of referential interpretations of a test, we can find different procedures performed both by Classical Test Theory (CTT) and the Theory of Item Response (IRT). Especially in this case (IRT), we can admit one test as a default, so to use its standardization and transfer the cut-off point to another instrument. Based on this information, the present study aimed to provide a cutoff score for the Baptista Depression Scale - Adult Version (EBADEP-A) through procedures of norms-transfer based on the Center for Epidemiologic Studies – Depression Scale (CES-D). The EBADEP-A presented good distribution and ability to discriminate depressive symptoms, and the sample, consisting of Brazilian College students, received a cutoff score of 32 points. It is emphasized that this is an exploratory and preliminary study, and it is suggested further analyzes to be performed with clinical samples for which results can be corroborated or confronted.

Using Trauma Injury Severity Score (TRISS) variables to predict length of hospital stay following trauma in New Zealand

Aim – To develop and assess the predictive capabilities of a statistical model that relates routinely collected Trauma Injury Severity Score (TRISS) variables to length of hospital stay (LOS) in survivors of traumatic injury. Method – Retrospective cohort study of adults who sustained a serious traumatic injury, and who survived until discharge from Auckland City, Middlemore, Waikato, or North Shore Hospitals between 2002 and 2006. Cubic-root transformed LOS was analysed using two-level mixed-effects regression models. Results – 1498 eligible patients were identified, 1446 (97%) injured from a blunt mechanism and 52 (3%) from a penetrating mechanism. For blunt mechanism trauma, 1096 (76%) were male, average age was 37 years (range: 15-94 years), and LOS and TRISS score information was available for 1362 patients. Spearman’s correlation and the median absolute prediction error between LOS and the original TRISS model was ρ=0.31 and 10.8 days, respectively, and between LOS and the final multivariable two-level mixed-effects regression model was ρ=0.38 and 6.0 days, respectively. Insufficient data were available for the analysis of penetrating mechanism models. Conclusions – Neither the original TRISS model nor the refined model has sufficient ability to accurately or reliably predict LOS. Additional predictor variables for LOS and other indicators for morbidity need to be considered.

Contemporary New Zealand coefficients for the Trauma Injury Severity Score: TRISS(NZ)

Aims – To develop local contemporary coefficients for the Trauma Injury Severity Score in New Zealand, TRISS(NZ), and to evaluate their performance at predicting survival against the original TRISS coefficients. Methods – Retrospective cohort study of adults who sustained a serious traumatic injury, and who survived until presentation at Auckland City, Middlemore, Waikato, or North Shore Hospitals between 2002 and 2006. Coefficients were estimated using ordinary and multilevel mixed-effects logistic regression models. Results – 1735 eligible patients were identified, 1672 (96%) injured from a blunt mechanism and 63 (4%) from a penetrating mechanism. For blunt mechanism trauma, 1250 (75%) were male and average age was 38 years (range: 15-94 years). TRISS information was available for 1565 patients of whom 204 (13%) died. Area under the Receiver Operating Characteristic (ROC) curves was 0.901 (95%CI: 0.879-0.923) for the TRISS(NZ) model and 0.890 (95% CI: 0.866-0.913) for TRISS (P<0.001). Insufficient data were available to determine coefficients for penetrating mechanism TRISS(NZ) models. Conclusions – Both TRISS models accurately predicted survival for blunt mechanism trauma. However, TRISS(NZ) coefficients were statistically superior to TRISS coefficients. A strong case exists for replacing TRISS coefficients in the New Zealand benchmarking software with these updated TRISS(NZ) estimates.

Data-driven impostor selection for T-norm score normalisation and the background dataset in SVM-based speaker verification

A data-driven background dataset refinement technique was recently proposed for SVM based speaker verification. This method selects a refined SVM background dataset from a set of candidate impostor examples after individually ranking examples by their relevance. This paper extends this technique to the refinement of the T-norm dataset for SVM-based speaker verification. The independent refinement of the background and T-norm datasets provides a means of investigating the sensitivity of SVM-based speaker verification performance to the selection of each of these datasets. Using refined datasets provided improvements of 13% in min. DCF and 9% in EER over the full set of impostor examples on the 2006 SRE corpus with the majority of these gains due to refinement of the T-norm dataset. Similar trends were observed for the unseen data of the NIST 2008 SRE.

Changing musical emotion : a computational rule system for modifying score and performance

When communicating emotion in music, composers and performers encode their expressive intentions through the control of basic musical features such as: pitch, loudness, timbre, mode, and articulation. The extent to which emotion can be controlled through the systematic manipulation of these features has not been fully examined. In this paper we present CMERS, a Computational Music Emotion Rule System for the control of perceived musical emotion that modifies features at the levels of score and performance in real-time. CMERS performance was evaluated in two rounds of perceptual testing. In experiment I, 20 participants continuously rated the perceived emotion of 15 music samples generated by CMERS. Three music works, each with five emotional variations were used (normal, happy, sad, angry, and tender). The intended emotion by CMERS was correctly identified 78% of the time, with significant shifts in valence and arousal also recorded, regardless of the works’ original emotion.

Trauma and Injury Severity Score (TRISS) coefficients 2009 revision

Background: Currently used Trauma and Injury Severity Score (TRISS) coefficients, which measure probability of survival (Ps), were derived from the Major Trauma Outcome Study (MTOS) in 1995 and are now unlikely to be optimal. This study aims to estimate new TRISS coefficients using a contemporary database of injured patients presenting to emergency departments in the United States; and to compare these against the MTOS coefficients.---------- Methods: Data were obtained from the National Trauma Data Bank (NTDB) and the NTDB National Sample Project (NSP). TRISS coefficients were estimated using logistic regression. Separate coefficients were derived from complete case and multistage multiple imputation analyses for each NTDB and NSP dataset. Associated Ps over Injury Severity Score values were graphed and compared by age (adult ≥ 15 years; pediatric < 15 years) and injury mechanism (blunt; penetrating) groups. Area under the Receiver Operating Characteristic curves was used to assess coefficients’ predictive performance.---------- Results: Overall 1,072,033 NTDB and 1,278,563 weighted NSP injury events were included, compared with 23,177 used in the original MTOS analyses. Large differences were seen between results from complete case and imputed analyses. For blunt mechanism and adult penetrating mechanism injuries, there were similarities between coefficients estimated on imputed samples, and marked divergences between associated Ps estimated and those from the MTOS. However, negligible differences existed between area under the receiver operating characteristic curves estimates because the overwhelming majority of patients had minor trauma and survived. For pediatric penetrating mechanism injuries, variability in coefficients was large and Ps estimates unreliable.---------- Conclusions: Imputed NTDB coefficients are recommended as the TRISS coefficients 2009 revision for blunt mechanism and adult penetrating mechanism injuries. Coefficients for pediatric penetrating mechanism injuries could not be reliably estimated.

Reflection on the use of rating and score cards for asset and service assessment

The concept of star rating council facilities has progressively gained traction in Australia following the work of Dean Taylor at Marochy Shire Council in Queensland in 2006 – 2007 and more recently by the Victorian STEP asset management program. The following paper provides a brief discussion on the use and merits of star rating within community asset management. We suggest that the current adoption of the star rating system to manage community investment in services is lacking in consistency. It is suggested that the major failing is a lack of clear understanding in the purpose being served by the systems. The discussion goes on to make some recommendations on how the concept of a star system could be further enhanced to serve the needs of our communities better.

Automatic spoken language identification utilizing acoustic and phonetic speech information

Automatic spoken Language Identi¯cation (LID) is the process of identifying the language spoken within an utterance. The challenge that this task presents is that no prior information is available indicating the content of the utterance or the identity of the speaker. The trend of globalization and the pervasive popularity of the Internet will amplify the need for the capabilities spoken language identi¯ca- tion systems provide. A prominent application arises in call centers dealing with speakers speaking di®erent languages. Another important application is to index or search huge speech data archives and corpora that contain multiple languages. The aim of this research is to develop techniques targeted at producing a fast and more accurate automatic spoken LID system compared to the previous National Institute of Standards and Technology (NIST) Language Recognition Evaluation. Acoustic and phonetic speech information are targeted as the most suitable fea- tures for representing the characteristics of a language. To model the acoustic speech features a Gaussian Mixture Model based approach is employed. Pho- netic speech information is extracted using existing speech recognition technol- ogy. Various techniques to improve LID accuracy are also studied. One approach examined is the employment of Vocal Tract Length Normalization to reduce the speech variation caused by di®erent speakers. A linear data fusion technique is adopted to combine the various aspects of information extracted from speech. As a result of this research, a LID system was implemented and presented for evaluation in the 2003 Language Recognition Evaluation conducted by the NIST.

Focus-score weighted super-resolution for uncooperative iris recognition at a distance and on the move

Uncooperative iris identification systems at a distance and on the move often suffer from poor resolution and poor focus of the captured iris images. The lack of pixel resolution and well-focused images significantly degrades the iris recognition performance. This paper proposes a new approach to incorporate the focus score into a reconstruction-based super-resolution process to generate a high resolution iris image from a low resolution and focus inconsistent video sequence of an eye. A reconstruction-based technique, which can incorporate middle and high frequency components from multiple low resolution frames into one desired super-resolved frame without introducing false high frequency components, is used. A new focus assessment approach is proposed for uncooperative iris at a distance and on the move to improve performance for variations in lighting, size and occlusion. A novel fusion scheme is then proposed to incorporate the proposed focus score into the super-resolution process. The experiments conducted on the The Multiple Biometric Grand Challenge portal database shows that our proposed approach achieves an EER of 2.1%, outperforming the existing state-of-the-art averaging signal-level fusion approach by 19.2% and the robust mean super-resolution approach by 8.7%.

A prospective pilot study evaluating the ‘cardiac decompensation score’ in the setting of intraaortic balloon counterpulsation

The study objective was to determine whether the ‘cardiac decompensation score’ could identify cardiac decompensation in a patient with existing cardiac compromise managed with intraaortic balloon counterpulsation (IABP). A one-group, posttest-only design was utilised to collect observations in 2003 from IABP recipients treated in the intensive care unit of a 450 bed Australian, government funded, public, cardiothoracic, tertiary referral hospital. Twenty-three consecutive IABP recipients were enrolled, four of whom died in ICU (17.4%). All non-survivors exhibited primarily rising scores over the observation period (p < 0.001) and had final scores of 25 or higher. In contrast, the maximum score obtained by a survivor at any time was 15. Regardless of survival, scores for the 23 participants were generally decreasing immediately following therapy escalation (p = 0.016). Further reflecting these changes in patient support, there was also a trend for scores to move from rising to falling at such treatment escalations (p = 0.024). This pilot study indicates the ‘cardiac decompensation score’ to accurately represent changes in heart function specific to an individual patient. Use of the score in conjunction with IABP may lead to earlier identification of changes occurring in a patient's cardiac function and thus facilitate improved IABP outcomes.