Cross likelihood ratio based speaker clustering using eigenvoice models

This paper proposes the use of eigenvoice modeling techniques with the Cross Likelihood Ratio (CLR) as a criterion for speaker clustering within a speaker diarization system. The CLR has previously been shown to be a robust decision criterion for speaker clustering using Gaussian Mixture Models. Recently, eigenvoice modeling techniques have become increasingly popular, due to its ability to adequately represent a speaker based on sparse training data, as well as an improved capture of differences in speaker characteristics. This paper hence proposes that it would be beneficial to capitalize on the advantages of eigenvoice modeling in a CLR framework. Results obtained on the 2002 Rich Transcription (RT-02) Evaluation dataset show an improved clustering performance, resulting in a 35.1% relative improvement in the overall Diarization Error Rate (DER) compared to the baseline system.

Eigenvoice modeling for cross likelihood ratio based speaker clustering : a Bayesian approach

This paper proposes the use of Bayesian approaches with the cross likelihood ratio (CLR) as a criterion for speaker clustering within a speaker diarization system, using eigenvoice modeling techniques. The CLR has previously been shown to be an effective decision criterion for speaker clustering using Gaussian mixture models. Recently, eigenvoice modeling has become an increasingly popular technique, due to its ability to adequately represent a speaker based on sparse training data, as well as to provide an improved capture of differences in speaker characteristics. The integration of eigenvoice modeling into the CLR framework to capitalize on the advantage of both techniques has also been shown to be beneficial for the speaker clustering task. Building on that success, this paper proposes the use of Bayesian methods to compute the conditional probabilities in computing the CLR, thus effectively combining the eigenvoice-CLR framework with the advantages of a Bayesian approach to the diarization problem. Results obtained on the 2002 Rich Transcription (RT-02) Evaluation dataset show an improved clustering performance, resulting in a 33.5% relative improvement in the overall Diarization Error Rate (DER) compared to the baseline system.

Extending the task of diarization to speaker attribution

In this paper we extend the concept of speaker annotation within a single-recording, or speaker diarization, to a collection wide approach we call speaker attribution. Accordingly, speaker attribution is the task of clustering expectantly homogenous intersession clusters obtained using diarization according to common cross-recording identities. The result of attribution is a collection of spoken audio across multiple recordings attributed to speaker identities. In this paper, an attribution system is proposed using mean-only MAP adaptation of a combined-gender UBM to model clusters from a perfect diarization system, as well as a JFA-based system with session variability compensation. The normalized cross-likelihood ratio is calculated for each pair of clusters to construct an attribution matrix and the complete linkage algorithm is employed to conduct clustering of the inter-session clusters. A matched cluster purity and coverage of 87.1% was obtained on the NIST 2008 SRE corpus.

Speaker linking using complete-linkage clustering

Speaker diarization determines instances of the same speaker within a recording. Extending this task to a collection of recordings for linking together segments spoken by a unique speaker requires speaker linking. In this paper we propose a speaker linking system using linkage clustering and state-of-the-art speaker recognition techniques. We evaluate our approach against two baseline linking systems using agglomerative cluster merging (AC) and agglomerative clustering with model retraining (ACR). We demonstrate that our linking method, using complete-linkage clustering, provides a relative improvement of 20% and 29% in attribution error rate (AER), over the AC and ACR systems, respectively.

Robust automatic speaker linking and attribution

This research makes a major contribution which enables efficient searching and indexing of large archives of spoken audio based on speaker identity. It introduces a novel technique dubbed as “speaker attribution” which is the task of automatically determining ‘who spoke when?’ in recordings and then automatically linking the unique speaker identities within each recording across multiple recordings. The outcome of the research will also have significant impact in improving the performance of automatic speech recognition systems through the extracted speaker identities.

Speaker attribution of Australian broadcast news data

Speaker attribution is the task of annotating a spoken audio archive based on speaker identities. This can be achieved using speaker diarization and speaker linking. In our previous work, we proposed an efficient attribution system, using complete-linkage clustering, for conducting attribution of large sets of two-speaker telephone data. In this paper, we build on our proposed approach to achieve a robust system, applicable to multiple recording domains. To do this, we first extend the diarization module of our system to accommodate multi-speaker (>2) recordings. We achieve this through using a robust cross-likelihood ratio (CLR) threshold stopping criterion for clustering, as opposed to the original stopping criterion of two speakers used for telephone data. We evaluate this baseline diarization module across a dataset of Australian broadcast news recordings, showing a significant lack of diarization accuracy without previous knowledge of the true number of speakers within a recording. We thus propose applying an additional pass of complete-linkage clustering to the diarization module, demonstrating an absolute improvement of 20% in diarization error rate (DER). We then evaluate our proposed multi-domain attribution system across the broadcast news data, demonstrating achievable attribution error rates (AER) as low as 17%.

A speaker rediarization scheme for improving diarization in large two-speaker telephone datasets

In this paper we propose a novel scheme for carrying out speaker diarization in an iterative manner. We aim to show that the information obtained through the first pass of speaker diarization can be reused to refine and improve the original diarization results. We call this technique speaker rediarization and demonstrate the practical application of our rediarization algorithm using a large archive of two-speaker telephone conversation recordings. We use the NIST 2008 SRE summed telephone corpora for evaluating our speaker rediarization system. This corpus contains recurring speaker identities across independent recording sessions that need to be linked across the entire corpus. We show that our speaker rediarization scheme can take advantage of inter-session speaker information, linked in the initial diarization pass, to achieve a 30% relative improvement over the original diarization error rate (DER) after only two iterations of rediarization.

Normality under uncertainty

Consider the demand for a good whose consumption be chosen prior to the resolution of uncertainty regarding income. How do changes in the distribution of income affect the demand for this good? In this paper we show that normality, is sufficient to guarantee that consumption increases of the Radon-Nikodym derivative of the new distribution with respect to the old is non-decreasing in the whole domain. However, if only first order stochastic dominance is assumed more structure must be imposed on preferences to guanantee the validity of the result. Finally a converse of the first result also obtains. If the change in measure is characterized by non-decreasing Radon-Nicodyn derivative, consumption of such a good will always increase if and only if the good is normal.

Cross-validated maximum likelihood enhances crystallographic simulated annealing refinement

Recently, the target function for crystallographic refinement has been improved through a maximum likelihood analysis, which makes proper allowance for the effects of data quality, model errors, and incompleteness. The maximum likelihood target reduces the significance of false local minima during the refinement process, but it does not completely eliminate them, necessitating the use of stochastic optimization methods such as simulated annealing for poor initial models. It is shown that the combination of maximum likelihood with cross-validation, which reduces overfitting, and simulated annealing by torsion angle molecular dynamics, which simplifies the conformational search problem, results in a major improvement of the radius of convergence of refinement and the accuracy of the refined structure. Torsion angle molecular dynamics and the maximum likelihood target function interact synergistically, the combination of both methods being significantly more powerful than each method individually. This is demonstrated in realistic test cases at two typical minimum Bragg spacings (dmin = 2.0 and 2.8 Å, respectively), illustrating the broad applicability of the combined method. In an application to the refinement of a new crystal structure, the combined method automatically corrected a mistraced loop in a poor initial model, moving the backbone by 4 Å.

Critical gap estimation by numerical and statistical highest likelihood search

Many traffic situations require drivers to cross or merge into a stream having higher priority. Gap acceptance theory enables us to model such processes to analyse traffic operation. This discussion demonstrated that numerical search fine tuned by statistical analysis can be used to determine the most likely critical gap for a sample of drivers, based on their largest rejected gap and accepted gap. This method shares some common features with the Maximum Likelihood Estimation technique (Troutbeck 1992) but lends itself well to contemporary analysis tools such as spreadsheet and is particularly analytically transparent. This method is considered not to bias estimation of critical gap due to very small rejected gaps or very large rejected gaps. However, it requires a sufficiently large sample that there is reasonable representation of largest rejected gap/accepted gap pairs within a fairly narrow highest likelihood search band.

Use of virtual index for measuring efficiency of innovation systems : a cross-country study

In this study we propose a virtual index for measuring the relative innovativeness of countries. Using a multistage virtual benchmarking process, the best and rational benchmark is extracted for inefficient ISs. Furthermore, Tobit and Ordinary Least Squares (OLS) regression models are used to investigate the likelihood of changes in inefficiencies by investigating country-specific factors. The empirical results relating to the virtual benchmarking process suggest that the OLS regression model would better explain changes in the performance of innovation- inefficient countries.

Burglar target selection : a cross-national comparison

Objectives This study builds on research undertaken by Bernasco and Nieuwbeerta and explores the generalizability of a theoretically derived offender target selection model in three cross-national study regions. Methods Taking a discrete spatial choice approach, we estimate the impact of both environment- and offender-level factors on residential burglary placement in the Netherlands, the United Kingdom, and Australia. Combining cleared burglary data from all study regions in a single statistical model, we make statistical comparisons between environments. Results In all three study regions, the likelihood an offender selects an area for burglary is positively influenced by proximity to their home, the proportion of easily accessible targets, and the total number of targets available. Furthermore, in two of the three study regions, juvenile offenders under the legal driving age are significantly more influenced by target proximity than adult offenders. Post hoc tests indicate the magnitudes of these impacts vary significantly between study regions. Conclusions While burglary target selection strategies are consistent with opportunity-based explanations of offending, the impact of environmental context is significant. As such, the approach undertaken in combining observations from multiple study regions may aid criminology scholars in assessing the generalizability of observed findings across multiple environments.

Back to normal: A retrospective, cross-sectional study of the multi-factorial determinants of normal birth in Queensland, Australia

Background Currently, care providers and policy-makers internationally are working to promote normal birth. In Australia, such initiatives are being implemented without any evidence of the prevalence or determinants of normal birth as a multidimensional construct. This study aimed to better understand the determinants of normal birth (defined as without induction of labour, epidural/spinal/general anaesthesia, forceps/vacuum, caesarean birth, or episiotomy) using secondary analyses of data from a population survey of women in Queensland, Australia. Methods Women who birthed in Queensland during a two-week period in 2009 were mailed a survey approximately three months after birth. Women (n=772) provided retrospective data on their pregnancy, labour and birth preferences and experiences, socio-demographic characteristics, and reproductive history. A series of logistic regressions were conducted to determine factors associated with having labour, having a vaginal birth, and having a normal birth. Findings Overall, 81.9% of women had labour, 66.4% had a vaginal birth, and 29.6% had a normal birth. After adjusting for other significant factors, women had significantly higher odds of having labour if they birthed in a public hospital and had a pre-existing preference for a vaginal birth. Of women who had labour, 80.8% had a vaginal birth. Women who had labour had significantly higher odds of having a vaginal birth if they attended antenatal classes, did not have continuous fetal monitoring, felt able to ‘take their time’ in labour, and had a pre-existing preference for a vaginal birth. Of women who had a vaginal birth, 44.7% had a normal birth. Women who had a vaginal birth had significantly higher odds of having a normal birth if they birthed in a public hospital, birthed outside regular business hours, had mobility in labour, did not have continuous fetal monitoring, and were non-supine during birth. Conclusions These findings provide a strong foundation on which to base resources aimed at increasing informed decision-making for maternity care consumers, providers, and policy-makers alike. Research to evaluate the impact of modifying key clinical practices (e.g., supporting women׳s mobility during labour, facilitating non-supine positioning during birth) on the likelihood of a normal birth is an important next step.

Responses in carcass lean pH at 24 hours post-mortem in lines of Large White pigs selected for growth rate on a fixed ration over a set time

A restricted maximum likelihood analysis applied to an animal model showed no significant differences (P > 0.05) in pH value of the longissimus dorsi measured at 24 h post-mortem (pH24) between high and low lines of Large White pigs selected over 4 years for post-weaning growth rate on restricted feeding. Genetic and phenotypic correlations between pH24 and production and carcass traits were estimated using all performance testing records combined with the pH24 measurements (5.05-7.02) on slaughtered animals. The estimate of heritability for pH24 was moderate (0.29 ± 0.18). Genetic correlations between pH24 and production or carcass composition traits, except for ultrasonic backfat (UBF), were not significantly different from zero. UBF had a moderate, positive genetic correlation with pH24 (0.24 ± 0.33). These estimates of genetic correlations affirmed that selection for increased growth rate on restricted feeding is likely to result in limited changes in pH24 and pork quality since the selection does not put a high emphasis on reduced fatness.

Measurement of the Single Top Quark Production Cross Section at CDF

We report a measurement of the single top quark production cross section in 2.2 ~fb-1 of p-pbar collision data collected by the Collider Detector at Fermilab at sqrt{s}=1.96 TeV. Candidate events are classified as signal-like by three parallel analyses which use likelihood, matrix element, and neural network discriminants. These results are combined in order to improve the sensitivity. We observe a signal consistent with the standard model prediction, but inconsistent with the background-only model by 3.7 standard deviations with a median expected sensitivity of 4.9 standard deviations. We measure a cross section of 2.2 +0.7 -0.6(stat+sys) pb, extract the CKM matrix element value |V_{tb}|=0.88 +0.13 -0.12 (stat+sys) +- 0.07(theory), and set the limit |V_{tb}|>0.66 at the 95% C.L.