For the last several years, speaker diarization has been attracting substantial research attention as one of the spoken language technologies applied for the improvement, or enrichment, of recording transcriptions. Recordings of meetings, compared to other domains, exhibit an increased complexity due to the spontaneity of speech, reverberation effects, and also due to the presence of overlapping speech. Overlapping speech refers to situations when two or more speakers are speaking simultaneously. In meeting data, a substantial portion of errors of the conventional speaker diarization systems can be ascribed to speaker overlaps, since usually only one speaker label is assigned per segment. Furthermore, simultaneous speech included in training data can eventually lead to corrupt single-speaker models and thus to a worse segmentation. This thesis concerns the detection of overlapping speech segments and its further application for the improvement of speaker diarization performance. We propose the use of three spatial cross-correlationbased parameters for overlap detection on distant microphone channel data. Spatial features from different microphone pairs are fused by means of principal component analysis, linear discriminant analysis, or by a multi-layer perceptron. In addition, we also investigate the possibility of employing longterm prosodic information. The most suitable subset from a set of candidate prosodic features is determined in two steps. Firstly, a ranking according to mRMR criterion is obtained, and then, a standard hill-climbing wrapper approach is applied in order to determine the optimal number of features. The novel spatial as well as prosodic parameters are used in combination with spectral-based features suggested previously in the literature. In experiments conducted on AMI meeting data, we show that the newly proposed features do contribute to the detection of overlapping speech, especially on data originating from a single recording site. In speaker diarization, for segments including detected speaker overlap, a second speaker label is picked, and such segments are also discarded from the model training. The proposed overlap labeling technique is integrated in Viterbi decoding, a part of the diarization algorithm. During the system development it was discovered that it is favorable to do an independent optimization of overlap exclusion and labeling with respect to the overlap detection system. We report improvements over the baseline diarization system on both single- and multi-site AMI data. Preliminary experiments with NIST RT data show DER improvement on the RT ¿09 meeting recordings as well. The addition of beamforming and TDOA feature stream into the baseline diarization system, which was aimed at improving the clustering process, results in a bit higher effectiveness of the overlap labeling algorithm. A more detailed analysis on the overlap exclusion behavior reveals big improvement contrasts between individual meeting recordings as well as between various settings of the overlap detection operation point. However, a high performance variability across different recordings is also typical of the baseline diarization system, without any overlap handling.

Victim offender overlap is a relatively new area of research with most studies having focused on applying this concept to the study of assault and homicide. Research in intimate partner violence has found that there exists a group of victim offenders or individuals who are involved in initiating, sustaining, and engaging in intimate partner violence as both victims and offenders (Johnson & Ferraro, 2000; Steinmetz, 1980). This Master’s thesis explored the concept of victim offender overlap in connection to intimate partner violence (IPV). Bivariate analyses were conducted using Paul C. Friday, Vivian Lord, M. Lyn Exum, and Jennifer L. Hartman’s (2003-2005) data, Evaluating the Impact of a Specialized Domestic Violence Police Unit in Charlotte, North Carolina. The findings suggest that there is a separate group of individuals involved in intimate partner violence who are both victims and offenders. Furthermore, the three groups (offenders only of IPV, victims only of IPV, and victim/offenders of IPV) were found to be different across gender, past criminal involvement, and future criminal activity as predicted. Suggestions for future research and the implications of the findings are discussed.

Joints represent a design challenge, especially for composite structures. Among the available joining methods, co-curing is an efficient way to integrate parts for some applications. Coates and Armanios have proposed a Single Nested Overlap (SNO) co-cured joint configuration, obtained from a single lap joint through the overlap/interleafing of the adjoining top/bottom adherend plies, respectively. Through a comparative investigation, they have demonstrated joint strength and fatigue life improvements over the single lap joint counterparts for unidirectional and quasi-isotropic adherend lay-ups. This research extends the comparative investigation of Coates and Armanios by focusing upon characterizing and differentiating the damage initiation and progression mechanisms under quasi-static loading. Six specimen configurations are manufactured and tested. It is confirmed that single nested overlap joints show 29.2% and 27.4% average improvement in strength over single lap counterparts for zero-degree unidirectional and quasi-isotropic lay-ups, respectively. Several nondestructive evaluation techniques are used to observe and analyze damage initiation, damage progression and failure modes of the studied specimens and to monitor their mechanical response. Using X-ray Radiography and Optical Microscopy techniques during quasi-static loading, a physical characterization of damage and failure mechanisms is obtained. The acoustic emission data acquired during monotonic loading could reveal the overall picture of AE activities produced by the damage initiation, development and accumulation mechanisms within the specimen via parametric analysis. Further AE analysis by a selected supervised clustering method is carried out and shown successful in differentiating and clustering the AE data. Correlation with physical observations from other techniques suggests that the resulting clusters may be associated to specific damage modes and failure mechanisms.

Cold laps are defined as micro-lack of fusion defects at the weld toe more or less parallel to the surface of the parent plate. These defects are known to negatively influence the fatigue properties of weldments. Previous studies suggest that cold lap formation can not be avoided completely in Gas Metal Arc Welding (GMAW). Therefore, a better understanding of formation mechanisms is imperative to be able to minimize the number and size of these defects. The main objective of this work has been to provide a more comprehensive understanding of cold laps, including categorising, characterisation and defining the most significant factors for formation. GMAW was used to produce welds that were investigated by metallographic methods using light optical microscopy, scanning electron microscopy and energy dispersive spectrometry. A novel classification of cold laps was introduced and three types of cold laps were identified: spatter cold laps, overlap cold laps and spatter-overlap cold laps. It was shown that cold laps are partially or fully filled by oxides. The most common oxides are manganese silicon oxides which were concluded to be formed primarily by oxidation of droplets. The presence of oxides was found to significantly increase the tendency to form spatter cold laps as well as overlap cold laps. Particularly for overlap cold laps, it was found that the depth (in transverse direction of weld) is reduced when welding in a non-oxidising environment. Welding on blasted surfaces increased the cold lap formation by entrapment of gas. The droplet and base metal temperatures were also found to be significant factors in cold lap formation. For overlap cold laps the occurrence frequency decreased with increased preheating temperature of the base metal. Mechanisms of overflowing resulting in overlap cold laps were discussed based on an extensive literature review. Several phenomena are believed to contribute to overflow including Rayleigh instability, the balance of forces, transfer of lateral momentum by droplets and an outward Marangoni fluid flow of the weld pool. The present studies suggest that cold lap formation can be suppressed by ensuring that the welding process (arc) is as stable as possible and by welding on a preheated work piece in a non-oxidising environment.

The biological diversity on planet Earth is astounding. Understanding the origins of this diversity, and how it is maintained, are the twin goals of ecology and evolutionary biology. An early and oft-repeated insight in this investigation is that that similar organisms cannot coexist indefinitely. Theory predicts that individuals and species will compete for limited resources and whichever has even a slight advantage will drive all others extinct in a process known as ‘competitive exclusion’. By diversifying, species avoid competition, thereby ‘stabilizing’ their coexistence. Yet natural systems often display levels of diversity that are surprisingly high, given this theory and investigations of how the similarity of coexisting species is maintained have received much less attention. Using a combination of field studies and experiments I demonstrate that highly similar species of freshwater amphipods may compete for resources without resulting in competitive exclusion. These findings suggest that there exist a range of interactions among Hyalella amphipods, ranging from strong stabilizing effects due to ecological trade-offs, to weakly stabilizing effects, to a total lack of stabilizing effects among various pairs of species in this system. These findings demonstrate how the relative strength of stabilizing forces may vary among coexisting species. Although much effort has been dedicated to enumerating and classifying the ways in which ecological and evolutionary forces promote diversity among species, there has been far less attention paid to mechanisms such as convergent evolution, habitat filtering, competition for non-substitutable resources, and non-ecological speciation, among others. I surveyed current theory that may explain the high levels of similarity among species often found in natural systems. I describe how several ecological and evolutionary mechanisms may operate to promote the coexistence of similar species and present results from new theoretical combinations of mechanisms to demonstrate how they may further act in concert with one another. / text

Background: Most traffic safety research projects require accurate real world data which is collected in different databases around the world. This is especially important since the results of these projects form the basis for new crash test procedures and standards. In many of these databases the involvement of the frontal structures of the car in frontal crashes is coded by using the SAE J224 practice (Society of Automobile Engineers). There were indications that by using this practice the database would contain an overestimate of the car frontal structure involvement in real world crashes. One purpose of this thesis is therefore to develop a new method for real world crash investigations to better address this issue. One purpose was also to adopt this method in a data collection of fatal crashes in Sweden and examine injury causation mechanisms. Studies shows that the commonly used Hybrid III dummy is not fully reproducing the kinematical behavior observed in frontal sled test with belted PMHS (Post Mortem Human Subject). A human FE-model (Finite Element) might be able to reproduce the behavior evidenced with the PMHS in order to study upper body kinematics in certain types of frontal collision events. Method: A new data collection method was developed with the purpose to examine actual load paths active in the car front during a frontal crash. An important purpose was to examine if there was a relation between these load paths and injury producing mechanisms. This was done in an examination and analysis of 61 fatally injured occupants in 53 car frontal crashes in a sample area covering 40 % of the population of Sweden. Sample period was one year (1st October 2000 to 30th September 2001). An existing human FE-model was developed and validated with respect to upper body kinematics by using existing frontal belted PMHS tests. This was done by building a FE-model of the seat and seat belt used in the PMHS tests. Results: A generic car structure was developed which was used in the data collection methodology. By adopting this new method, Small Overlap (SO) crashes emerged as the most common crash configuration (48 %) among belted frontal fatalities. The injury producing mechanism in SO crashes is characterized by occupant upper body impacts in the side structure (door, a-pillar) of the car. This upper body kinematics is induced by both the crash pulse and the asymmetrical three point belt system. Current crash test procedures are not designed to fully estimate the performance of neither car structures nor restraints in SO crashes. In order to develop a better tool for reproducing this kinematical behavior a FE-model of a human body was refined and validated for belted conditions. This validation was performed with satisfying result. Conclusions: This study showed that by adopting new methods of data collecting new areas of traffic safety could be considered. In this study SO (48 %) crashes emerged as the most common crash configuration for belted frontal fatalities. Approximately ¼ of the fatalities occurred in a crash configuration comparable to current barrier crash test procedures. The body kinematics of PMHS in the SO crashes can be replicated and studied by using a FE-model of a human body in the collision load case model. With this tool possible collision counter measures could be evaluated for the SO crash configuration.

The aim with this study was to develop a method to estimate total area of hair lichens and to compare morphological traits and water storage in them. Hair lichens are an important component of the epiphytic flora in boreal forests. Their growth is primarily regulated by available water, and light when hydrated. Lichens have no active mechanism to regulate their water content and their water holding capacity (WHC, mg H2O/cm2) is thus an important factor for how long they remain wet and metabolically active. In this study, the water uptake and loss in five hair lichens (Alectoria sarmentosa, three Bryoria spp. and Usnea dasypoga) were compared. Their area were estimated by combining photography, scanning and a computer programme that estimates the area of objects. Total area overlap of individual branches was calculated for each species, to estimate total area of the lichen. WHC and specific thallus mass (STM) (mg DM/cm2) of the lichens were calculated. Bryoria spp. had a significantly lower STM compared to U. dasypoga and A. sarmentosa, due to its thinner branches and higher branch density. Bryoria also had a lower WHC compared to A. sarmentosa, promoting a rapid uptake and loss of water. All species had a significant relationship between STM and WHC, above a 1:1 line for all species except U. dasypoga. The lower relationship in U. dasypoga is explained by its less developed branching in combination with its thick branches.

Consumer researchers have long accepted that consumption can serve as a tool in the creation, maintenance, and expression of consumers' identities. And brands, in particular, may be important to the consumer self, even potentially serving as relationship partners. This dissertation explores how brands that are important to consumers may impact their identities at a cognitive level. Specifically, I apply Aron et al.'s (1991) "including others in the self" paradigm from interpersonal relationship research, which finds that people's cognitive representations of close others overlap the psychological self, to studying the impact of brands on the self. I provide evidence that consumers' mental representations of loved brands, which may be perceived as relationship partners, overlap the psychological self in memory. I refer to this as self-brand overlap. I also consider the relevance of disliked brands to the consumer self, providing evidence that consumers' mental representations of disliked brands are dissociated from the psychological self in memory. I refer to this as self-brand dissociation.In two studies I demonstrate and replicate the self-brand overlap and dissociation effects. And, study 2 further explores these constructs, providing evidence that self-brand overlap and dissociation are the cognitive representations of positive and negative, respectively, consumer-brand relationships. This dissertation also includes three further studies that aim to identify boundary conditions of these effects. However, no conclusive support is found for a role of any of the explored moderators. Specifically, studies 3 through 5 find the self-brand dissociation effect in every condition, in every study, suggesting that self-brand dissociation is impervious to the boundary conditions examined. Evidence for the self-brand overlap effect, which was demonstrated in both studies 1 and 2, however, is mixed in each of these 3 later studies. Potential reasons for this lack of concrete replication are offered.

The genetic structure of an intra-host viral population has an effect on many clinically important phenotypic traits such as escape from vaccine induced immunity, virulence, and response to antiviral therapies. Next-generation sequencing provides read-coverage sufficient for genomic reconstruction of a heterogeneous, yet highly similar, viral population; and more specifically, for the detection of rare variants. Admittedly, while depth is less of an issue for modern sequencers, the short length of generated reads complicates viral population assembly. This task is worsened by the presence of both random and systematic sequencing errors in huge amounts of data. In this dissertation I present completed work for reconstructing a viral population given next-generation sequencing data. Several algorithms are described for solving this problem under the error-free amplicon (or sliding-window) model. In order for these methods to handle actual real-world data, an error-correction method is proposed. A formal derivation of its likelihood model along with optimization steps for an EM algorithm are presented. Although these methods perform well, they cannot take into account paired-end sequencing data. In order to address this, a new method is detailed that works under the error-free paired-end case along with maximum a-posteriori estimation of the model parameters.

Prison Animal Programs that bring inmates and dogs together consistently report improvements to inmates’ self-esteem, ability to empathize, and helping behaviors with no understanding of why these improvements occur. With similar improvement documented in relationship closeness literature, this research examines the felt inmate dog connection and self-reported closeness as a possible explanation for the three reported benefits. Introducing relationship closeness scales that substitute a dog for another person, 37 inmate handlers at three correctional facilities completed survey and interview questions measuring self-esteem, self-expansion, and self-other (dog) overlap.Research findings strongly support that inmate-dog connectivity is consistent with documented dyadic human connectivity to close family and friends. The self-reported connectivity between handler and dog is furthermore suggested to strongly influence the three areas of improvement with additional research recommended. Findings also suggest that working with dogs in prison initiates a healing or restorative process for both inmate handlers and the outside community.