Speaker diarization: Segmentation and clustering of speeches

Speaker diarization is the process of sorting speeches according to the speaker. Diarization helps to search and retrieve what a certain speaker uttered in a meeting. Applications of diarization systemsextend to other domains than meetings, for example, lectures, telephone, television, and radio. Besides, diarization enhances the performance of several speech technologies such as speaker recognition, automatic transcription, and speaker tracking. Methodologies previously used in developing diarization systems are discussed. Prior results and techniques are studied and compared. Methods such as Hidden Markov Models and Gaussian Mixture Models that are used in speaker recognition and other speech technologies are also used in speaker diarization. The objective of this thesis is to develop a speaker diarization system in meeting domain. Experimental part of this work indicates that zero-crossing rate can be used effectively in breaking down the audio stream into segments, and adaptive Gaussian Models fit adequately short audio segments. Results show that 35 Gaussian Models and one second as average length of each segment are optimum values to build a diarization system for the tested data. Uniting the segments which are uttered by same speaker is done in a bottom-up clustering by a newapproach of categorizing the mixture weights.

Identification of croplands of winter cereals in Rio Grande do Sul state, Brazil, through unsupervised classification of normalized difference vegetation index images

This study aimed to propose methods to identify croplands cultivated with winter cereals in the northern region of Rio Grande do Sul State, Brazil. Thus, temporal profiles of Normalized Difference Vegetation Index (NDVI) from MODIS sensor, from April to December of the 2000 to 2008, were analyzed. Firstly, crop masks were elaborated by subtracting the minimum NDVI image (April to May) from the maximum NDVI image (June to October). Then, an unsupervised classification of NDVI images was carried out (Isodata), considering the crop mask areas. According to the results, crop masks allowed the identification of pixels with greatest green biomass variation. This variation might be associated or not with winter cereals areas established to grain production. The unsupervised classification generated classes in which NDVI temporal profiles were associated with water bodies, pastures, winter cereals for grain production and for soil cover. Temporal NDVI profiles of the class winter cereals for grain production were in agree with crop patterns in the region (developmental stage, management standard and sowing dates). Therefore, unsupervised classification based on crop masks allows distinguishing and monitoring winter cereal crops, which were similar in terms of morphology and phenology.

Management zones using fuzzy clustering based on spatial-temporal variability of soil and corn yield

Clustering soil and crop data can be used as a basis for the definition of management zones because the data are grouped into clusters based on the similar interaction of these variables. Therefore, the objective of this study was to identify management zones using fuzzy c-means clustering analysis based on the spatial and temporal variability of soil attributes and corn yield. The study site (18 by 250-m in size) was located in Jaboticabal, São Paulo/Brazil. Corn yield was measured in one hundred 4.5 by 10-m cells along four parallel transects (25 observations per transect) over five growing seasons between 2001 and 2010. Soil chemical and physical attributes were measured. SAS procedure MIXED was used to identify which variable(s) most influenced the spatial variability of corn yield over the five study years. Basis saturation (BS) was the variable that better related to corn yield, thus, semivariograms models were fitted for BS and corn yield and then, data values were krigged. Management Zone Analyst software was used to carry out the fuzzy c-means clustering algorithm. The optimum number of management zones can change over time, as well as the degree of agreement between the BS and corn yield management zone maps. Thus, it is very important take into account the temporal variability of crop yield and soil attributes to delineate management zones accurately.

Online peer to peer lending: clustering borrowers using self-organizing maps

The purpose of this thesis is to find out whether all the peer to peer lenders are unworthy of credit and also if there are single qualities or combinations of qualities that determine the probability of default of a person or group of people. Distinguishing qualities are searched with self-organizing maps (SOM). Qualities and groups of people found by the self-organizing map are then compared to the average. The comparison is carried out by looking how big proportion of borrowers meeting the criteria is two months or more behind with their payments. Research data used is collected by an Estonian peer to peer lending company during the years of 2011-2014. Data consists of peer to peer borrowers and information gathered from them.

Effect of race, genetic population structure, and genetic models in two-locus association studies: clustering of functional renin-angiotensin system gene variants in hypertension association studies

Previous genetic association studies have overlooked the potential for biased results when analyzing different population structures in ethnically diverse populations. The purpose of the present study was to quantify this bias in two-locus association studies conducted on an admixtured urban population. We studied the genetic structure distribution of angiotensin-converting enzyme insertion/deletion (ACE I/D) and angiotensinogen methionine/threonine (M/T) polymorphisms in 382 subjects from three subgroups in a highly admixtured urban population. Group I included 150 white subjects; group II, 142 mulatto subjects, and group III, 90 black subjects. We conducted sample size simulation studies using these data in different genetic models of gene action and interaction and used genetic distance calculation algorithms to help determine the population structure for the studied loci. Our results showed a statistically different population structure distribution of both ACE I/D (P = 0.02, OR = 1.56, 95% CI = 1.05-2.33 for the D allele, white versus black subgroup) and angiotensinogen M/T polymorphism (P = 0.007, OR = 1.71, 95% CI = 1.14-2.58 for the T allele, white versus black subgroup). Different sample sizes are predicted to be determinant of the power to detect a given genotypic association with a particular phenotype when conducting two-locus association studies in admixtured populations. In addition, the postulated genetic model is also a major determinant of the power to detect any association in a given sample size. The present simulation study helped to demonstrate the complex interrelation among ethnicity, power of the association, and the postulated genetic model of action of a particular allele in the context of clustering studies. This information is essential for the correct planning and interpretation of future association studies conducted on this population.

Fuzzy tolerance/equivalence relations from intersection, union and complement operators and their application to a clustering method

This master thesis work introduces the fuzzy tolerance/equivalence relation and its application in cluster analysis. The work presents about the construction of fuzzy equivalence relations using increasing generators. Here, we investigate and research on the role of increasing generators for the creation of intersection, union and complement operators. The objective is to develop different varieties of fuzzy tolerance/equivalence relations using different varieties of increasing generators. At last, we perform a comparative study with these developed varieties of fuzzy tolerance/equivalence relations in their application to a clustering method.

Category fluency test: effects of age, gender and education on total scores, clustering and switching in Brazilian Portuguese-speaking subjects

Verbal fluency tests are used as a measure of executive functions and language, and can also be used to evaluate semantic memory. We analyzed the influence of education, gender and age on scores in a verbal fluency test using the animal category, and on number of categories, clustering and switching. We examined 257 healthy participants (152 females and 105 males) with a mean age of 49.42 years (SD = 15.75) and having a mean educational level of 5.58 (SD = 4.25) years. We asked them to name as many animals as they could. Analysis of variance was performed to determine the effect of demographic variables. No significant effect of gender was observed for any of the measures. However, age seemed to influence the number of category changes, as expected for a sensitive frontal measure, after being controlled for the effect of education. Educational level had a statistically significant effect on all measures, except for clustering. Subject performance (mean number of animals named) according to schooling was: illiterates, 12.1; 1 to 4 years, 12.3; 5 to 8 years, 14.0; 9 to 11 years, 16.7, and more than 11 years, 17.8. We observed a decrease in performance in these five educational groups over time (more items recalled during the first 15 s, followed by a progressive reduction until the fourth interval). We conclude that education had the greatest effect on the category fluency test in this Brazilian sample. Therefore, we must take care in evaluating performance in lower educational subjects.

Clustering of psychiatric and somatic illnesses in the general population: multimorbidity and socioeconomic correlates

The distribution of psychiatric disorders and of chronic medical illnesses was studied in a population-based sample to determine whether these conditions co-occur in the same individual. A representative sample (N = 1464) of adults living in households was assessed by the Composite International Diagnostic Interview, version 1.1, as part of the São Paulo Epidemiological Catchment Area Study. The association of sociodemographic variables and psychological symptoms regarding medical illness multimorbidity (8 lifetime somatic conditions) and psychiatric multimorbidity (15 lifetime psychiatric disorders) was determined by negative binomial regression. A total of 1785 chronic medical conditions and 1163 psychiatric conditions were detected in the population concentrated in 34.1 and 20% of respondents, respectively. Subjects reporting more psychiatric disorders had more medical illnesses. Characteristics such as age range (35-59 years, risk ratio (RR) = 1.3, and more than 60 years, RR = 1.7), being separated (RR = 1.2), being a student (protective effect, RR = 0.7), being of low educational level (RR = 1.2) and being psychologically distressed (RR = 1.1) were determinants of medical conditions. Age (35-59 years, RR = 1.2, and more than 60 years, RR = 0.5), being retired (RR = 2.5), and being psychologically distressed (females, RR = 1.5, and males, RR = 1.4) were determinants of psychiatric disorders. In conclusion, psychological distress and some sociodemographic features such as age, marital status, occupational status, educational level, and gender are associated with psychiatric and medical multimorbidity. The distribution of both types of morbidity suggests the need of integrating mental health into general clinical settings.

Hyperspectral Mineral Identification using SVM and SOM

Remote sensing techniques involving hyperspectral imagery have applications in a number of sciences that study some aspects of the surface of the planet. The analysis of hyperspectral images is complex because of the large amount of information involved and the noise within that data. Investigating images with regard to identify minerals, rocks, vegetation and other materials is an application of hyperspectral remote sensing in the earth sciences. This thesis evaluates the performance of two classification and clustering techniques on hyperspectral images for mineral identification. Support Vector Machines (SVM) and Self-Organizing Maps (SOM) are applied as classification and clustering techniques, respectively. Principal Component Analysis (PCA) is used to prepare the data to be analyzed. The purpose of using PCA is to reduce the amount of data that needs to be processed by identifying the most important components within the data. A well-studied dataset from Cuprite, Nevada and a dataset of more complex data from Baffin Island were used to assess the performance of these techniques. The main goal of this research study is to evaluate the advantage of training a classifier based on a small amount of data compared to an unsupervised method. Determining the effect of feature extraction on the accuracy of the clustering and classification method is another goal of this research. This thesis concludes that using PCA increases the learning accuracy, and especially so in classification. SVM classifies Cuprite data with a high precision and the SOM challenges SVM on datasets with high level of noise (like Baffin Island).

Assessing the Reproducibility of Clustering of Molecular Dynamics Conformations on Self-Organizing Maps

The goal of most clustering algorithms is to find the optimal number of clusters (i.e. fewest number of clusters). However, analysis of molecular conformations of biological macromolecules obtained from computer simulations may benefit from a larger array of clusters. The Self-Organizing Map (SOM) clustering method has the advantage of generating large numbers of clusters, but often gives ambiguous results. In this work, SOMs have been shown to be reproducible when the same conformational dataset is independently clustered multiple times (~100), with the help of the Cramérs V-index (C_v). The ability of C_v to determine which SOMs are reproduced is generalizable across different SOM source codes. The conformational ensembles produced from MD (molecular dynamics) and REMD (replica exchange molecular dynamics) simulations of the penta peptide Met-enkephalin (MET) and the 34 amino acid protein human Parathyroid Hormone (hPTH) were used to evaluate SOM reproducibility. The training length for the SOM has a huge impact on the reproducibility. Analysis of MET conformational data definitively determined that toroidal SOMs cluster data better than bordered maps due to the fact that toroidal maps do not have an edge effect. For the source code from MATLAB, it was determined that the learning rate function should be LINEAR with an initial learning rate factor of 0.05 and the SOM should be trained by a sequential algorithm. The trained SOMs can be used as a supervised classification for another dataset. The toroidal 10×10 hexagonal SOMs produced from the MATLAB program for hPTH conformational data produced three sets of reproducible clusters (27%, 15%, and 13% of 100 independent runs) which find similar partitionings to those of smaller 6×6 SOMs. The χ^2 values produced as part of the C_v calculation were used to locate clusters with identical conformational memberships on independently trained SOMs, even those with different dimensions. The χ^2 values could relate the different SOM partitionings to each other.

Understanding deep architectures and the effect of unsupervised pre-training

Cette thèse porte sur une classe d'algorithmes d'apprentissage appelés architectures profondes. Il existe des résultats qui indiquent que les représentations peu profondes et locales ne sont pas suffisantes pour la modélisation des fonctions comportant plusieurs facteurs de variation. Nous sommes particulièrement intéressés par ce genre de données car nous espérons qu'un agent intelligent sera en mesure d'apprendre à les modéliser automatiquement; l'hypothèse est que les architectures profondes sont mieux adaptées pour les modéliser. Les travaux de Hinton (2006) furent une véritable percée, car l'idée d'utiliser un algorithme d'apprentissage non-supervisé, les machines de Boltzmann restreintes, pour l'initialisation des poids d'un réseau de neurones supervisé a été cruciale pour entraîner l'architecture profonde la plus populaire, soit les réseaux de neurones artificiels avec des poids totalement connectés. Cette idée a été reprise et reproduite avec succès dans plusieurs contextes et avec une variété de modèles. Dans le cadre de cette thèse, nous considérons les architectures profondes comme des biais inductifs. Ces biais sont représentés non seulement par les modèles eux-mêmes, mais aussi par les méthodes d'entraînement qui sont souvent utilisés en conjonction avec ceux-ci. Nous désirons définir les raisons pour lesquelles cette classe de fonctions généralise bien, les situations auxquelles ces fonctions pourront être appliquées, ainsi que les descriptions qualitatives de telles fonctions. L'objectif de cette thèse est d'obtenir une meilleure compréhension du succès des architectures profondes. Dans le premier article, nous testons la concordance entre nos intuitions---que les réseaux profonds sont nécessaires pour mieux apprendre avec des données comportant plusieurs facteurs de variation---et les résultats empiriques. Le second article est une étude approfondie de la question: pourquoi l'apprentissage non-supervisé aide à mieux généraliser dans un réseau profond? Nous explorons et évaluons plusieurs hypothèses tentant d'élucider le fonctionnement de ces modèles. Finalement, le troisième article cherche à définir de façon qualitative les fonctions modélisées par un réseau profond. Ces visualisations facilitent l'interprétation des représentations et invariances modélisées par une architecture profonde.

Clustering algorithms and shape factor methods to discriminate among small GTPase phenotypes using DIC image analysis.

Naïvement perçu, le processus d’évolution est une succession d’événements de duplication et de mutations graduelles dans le génome qui mènent à des changements dans les fonctions et les interactions du protéome. La famille des hydrolases de guanosine triphosphate (GTPases) similaire à Ras constitue un bon modèle de travail afin de comprendre ce phénomène fondamental, car cette famille de protéines contient un nombre limité d’éléments qui diffèrent en fonctionnalité et en interactions. Globalement, nous désirons comprendre comment les mutations singulières au niveau des GTPases affectent la morphologie des cellules ainsi que leur degré d’impact sur les populations asynchrones. Mon travail de maîtrise vise à classifier de manière significative différents phénotypes de la levure Saccaromyces cerevisiae via l’analyse de plusieurs critères morphologiques de souches exprimant des GTPases mutées et natives. Notre approche à base de microscopie et d’analyses bioinformatique des images DIC (microscopie d’interférence différentielle de contraste) permet de distinguer les phénotypes propres aux cellules natives et aux mutants. L’emploi de cette méthode a permis une détection automatisée et une caractérisation des phénotypes mutants associés à la sur-expression de GTPases constitutivement actives. Les mutants de GTPases constitutivement actifs Cdc42 Q61L, Rho5 Q91H, Ras1 Q68L et Rsr1 G12V ont été analysés avec succès. En effet, l’implémentation de différents algorithmes de partitionnement, permet d’analyser des données qui combinent les mesures morphologiques de population native et mutantes. Nos résultats démontrent que l’algorithme Fuzzy C-Means performe un partitionnement efficace des cellules natives ou mutantes, où les différents types de cellules sont classifiés en fonction de plusieurs facteurs de formes cellulaires obtenus à partir des images DIC. Cette analyse démontre que les mutations Cdc42 Q61L, Rho5 Q91H, Ras1 Q68L et Rsr1 G12V induisent respectivement des phénotypes amorphe, allongé, rond et large qui sont représentés par des vecteurs de facteurs de forme distincts. Ces distinctions sont observées avec différentes proportions (morphologie mutante / morphologie native) dans les populations de mutants. Le développement de nouvelles méthodes automatisées d’analyse morphologique des cellules natives et mutantes s’avère extrêmement utile pour l’étude de la famille des GTPases ainsi que des résidus spécifiques qui dictent leurs fonctions et réseau d’interaction. Nous pouvons maintenant envisager de produire des mutants de GTPases qui inversent leur fonction en ciblant des résidus divergents. La substitution fonctionnelle est ensuite détectée au niveau morphologique grâce à notre nouvelle stratégie quantitative. Ce type d’analyse peut également être transposé à d’autres familles de protéines et contribuer de manière significative au domaine de la biologie évolutive.

SPATIAL CLUSTERING ALGORITHMS-AN OVERVIEW.

An Overview of known spatial clustering algorithms The space of interest can be the two-dimensional abstraction of the surface of the earth or a man-made space like the layout of a VLSI design, a volume containing a model of the human brain, or another 3d-space representing the arrangement of chains of protein molecules. The data consists of geometric information and can be either discrete or continuous. The explicit location and extension of spatial objects define implicit relations of spatial neighborhood (such as topological, distance and direction relations) which are used by spatial data mining algorithms. Therefore, spatial data mining algorithms are required for spatial characterization and spatial trend analysis. Spatial data mining or knowledge discovery in spatial databases differs from regular data mining in analogous with the differences between non-spatial data and spatial data. The attributes of a spatial object stored in a database may be affected by the attributes of the spatial neighbors of that object. In addition, spatial location, and implicit information about the location of an object, may be exactly the information that can be extracted through spatial data mining

Detection of Flock Movement in Spatio-Temporal Database using Clustering Techniques - an Experience

In this paper, moving flock patterns are mined from spatio- temporal datasets by incorporating a clustering algorithm. A flock is defined as the set of data that move together for a certain continuous amount of time. Finding out moving flock patterns using clustering algorithms is a potential method to find out frequent patterns of movement in large trajectory datasets. In this approach, SPatial clusteRing algoRithm thrOugh sWarm intelligence (SPARROW) is the clustering algorithm used. The advantage of using SPARROW algorithm is that it can effectively discover clusters of widely varying sizes and shapes from large databases. Variations of the proposed method are addressed and also the experimental results show that the problem of scalability and duplicate pattern formation is addressed. This method also reduces the number of patterns produced

Spectral clustering independent component analysis for tissue classification from brain MRI

A spectral angle based feature extraction method, Spectral Clustering Independent Component Analysis (SC-ICA), is proposed in this work to improve the brain tissue classification from Magnetic Resonance Images (MRI). SC-ICA provides equal priority to global and local features; thereby it tries to resolve the inefficiency of conventional approaches in abnormal tissue extraction. First, input multispectral MRI is divided into different clusters by a spectral distance based clustering. Then, Independent Component Analysis (ICA) is applied on the clustered data, in conjunction with Support Vector Machines (SVM) for brain tissue analysis. Normal and abnormal datasets, consisting of real and synthetic T1-weighted, T2-weighted and proton density/fluid-attenuated inversion recovery images, were used to evaluate the performance of the new method. Comparative analysis with ICA based SVM and other conventional classifiers established the stability and efficiency of SC-ICA based classification, especially in reproduction of small abnormalities. Clinical abnormal case analysis demonstrated it through the highest Tanimoto Index/accuracy values, 0.75/98.8%, observed against ICA based SVM results, 0.17/96.1%, for reproduced lesions. Experimental results recommend the proposed method as a promising approach in clinical and pathological studies of brain diseases