New methodology for the integration of biometric features in speaker recognition systems applied to security environments

La cuestión principal abordada en esta tesis doctoral es la mejora de los sistemas biométricos de reconocimiento de personas a partir de la voz, proponiendo el uso de una nueva parametrización, que hemos denominado parametrización biométrica extendida dependiente de género (GDEBP en sus siglas en inglés). No se propone una ruptura completa respecto a los parámetros clásicos sino una nueva forma de utilizarlos y complementarlos. En concreto, proponemos el uso de parámetros diferentes dependiendo del género del locutor, ya que como es bien sabido, la voz masculina y femenina presentan características diferentes que deberán modelarse, por tanto, de diferente manera. Además complementamos los parámetros clásicos utilizados (MFFC extraídos de la señal de voz), con un nuevo conjunto de parámetros extraídos a partir de la deconstrucción de la señal de voz en sus componentes de fuente glótica (más relacionada con el proceso y órganos de fonación y por tanto con características físicas del locutor) y de tracto vocal (más relacionada con la articulación acústica y por tanto con el mensaje emitido). Para verificar la validez de esta propuesta se plantean diversos escenarios, utilizando diferentes bases de datos, para validar que la GDEBP permite generar una descripción más precisa de los locutores que los parámetros MFCC clásicos independientes del género. En concreto se plantean diferentes escenarios de identificación sobre texto restringido y texto independiente utilizando las bases de datos de HESPERIA y ALBAYZIN. El trabajo también se completa con la participación en dos competiciones internacionales de reconocimiento de locutor, NIST SRE (2010 y 2012) y MOBIO 2013. En el primer caso debido a la naturaleza de las bases de datos utilizadas se obtuvieron resultados cercanos al estado del arte, mientras que en el segundo de los casos el sistema presentado obtuvo la mejor tasa de reconocimiento para locutores femeninos. A pesar de que el objetivo principal de esta tesis no es el estudio de sistemas de clasificación, sí ha sido necesario analizar el rendimiento de diferentes sistemas de clasificación, para ver el rendimiento de la parametrización propuesta. En concreto, se ha abordado el uso de sistemas de reconocimiento basados en el paradigma GMM-UBM, supervectores e i-vectors. Los resultados que se presentan confirman que la utilización de características que permitan describir los locutores de manera más precisa es en cierto modo más importante que la elección del sistema de clasificación utilizado por el sistema. En este sentido la parametrización propuesta supone un paso adelante en la mejora de los sistemas de reconocimiento biométrico de personas por la voz, ya que incluso con sistemas de clasificación relativamente simples se consiguen tasas de reconocimiento realmente competitivas. ABSTRACT The main question addressed in this thesis is the improvement of automatic speaker recognition systems, by the introduction of a new front-end module that we have called Gender Dependent Extended Biometric Parameterisation (GDEBP). This front-end do not constitute a complete break with respect to classical parameterisation techniques used in speaker recognition but a new way to obtain these parameters while introducing some complementary ones. Specifically, we propose a gender-dependent parameterisation, since as it is well known male and female voices have different characteristic, and therefore the use of different parameters to model these distinguishing characteristics should provide a better characterisation of speakers. Additionally, we propose the introduction of a new set of biometric parameters extracted from the components which result from the deconstruction of the voice into its glottal source estimate (close related to the phonation process and the involved organs, and therefore the physical characteristics of the speaker) and vocal tract estimate (close related to acoustic articulation and therefore to the spoken message). These biometric parameters constitute a complement to the classical MFCC extracted from the power spectral density of speech as a whole. In order to check the validity of this proposal we establish different practical scenarios, using different databases, so we can conclude that a GDEBP generates a more accurate description of speakers than classical approaches based on gender-independent MFCC. Specifically, we propose scenarios based on text-constrain and text-independent test using HESPERIA and ALBAYZIN databases. This work is also completed with the participation in two international speaker recognition evaluations: NIST SRE (2010 and 2012) and MOBIO 2013, with diverse results. In the first case, due to the nature of the NIST databases, we obtain results closed to state-of-the-art although confirming our hypothesis, whereas in the MOBIO SRE we obtain the best simple system performance for female speakers. Although the study of classification systems is beyond the scope of this thesis, we found it necessary to analise the performance of different classification systems, in order to verify the effect of them on the propose parameterisation. In particular, we have addressed the use of speaker recognition systems based on the GMM-UBM paradigm, supervectors and i-vectors. The presented results confirm that the selection of a set of parameters that allows for a more accurate description of the speakers is as important as the selection of the classification method used by the biometric system. In this sense, the proposed parameterisation constitutes a step forward in improving speaker recognition systems, since even when using relatively simple classification systems, really competitive recognition rates are achieved.

Numerical Investigation of the Flow Features around Heave Plates Oscillating Close to a Free Surface or Seabed

Performance of heave plates used in offshore structures is strongly influenced by their added mass and damping, which are affected by proximity to a boundary. A previous paper by the authors presented numerical simulations of the flow around a circular solid disk oscillating at varying elevations from seabed [1]. The force calculated was used to evaluate the added mass and damping coefficients for the disk. The simulations suggest that as the structure moves closer to the seabed the added mass and damping coefficients (Ca and Cb) increases continuously. In order to understand the physics behind the added mass and damping trends, when a heave plate is moving near a seabed or closer to the free surface, the flow characteristics around the heave plate are examined numerically in this paper. Flow around oscillating disks is dominated by generation and development of phase-dependent vortical structures, characterized by the KC number and the distance from the seabed or free surface to the heave plate. Numerical calculations presented in this paper have comprised the qualitative analysis of the vortex shedding and the investigation of the links between such vortex shedding and, on one hand the damping coefficient, and on the other hand, pairing mechanisms such as the shedding angle.

Greenfield shopping center development. The integration between environmental features, building design and investment analysis at the decision making stage

The Shopping centre is a long term investment in which Greenfield development decisions are often taken based on risks analysis regarding construction costs, location, competition, market and an expected DCF. Furthermore, integration between the building design, project planning, operational costs and investment analysis is not entirely considered by the investor at the decision making stage. The absence of such information tends to produce certain negative impacts on the future running costs and annual maintenance of the building, especially on energy demand and other occupancy expenses paid by the tenants to the landlord. From the investor´s point of view, this blind spot in strategy development will possibly decrease their profit margin as changes in the occupancy expenses[ ] have a direct outcome on the profit margin. In order to try to reduce some higher operating cost components such as energy use and other utility savings as well as their CO2 emissions, quite a few income properties worldwide have some type of environmental label such as BREEAM and LEED. The drawback identified in this labelling is that usually the investments required to get an ecolabel are high and the investor finds no direct evidence that it increases market value. However there is research on certified commercial properties (especially offices) that shows better performance in terms of occupancy rate and rental cost (Warren-Myers, 2012). Additionally, Sayce (2013) says that the certification only provides a quick reference point i.e. the lack of a certificate does not indicate that a building is not sustainable or efficient. Based on the issues described above, this research compares important components of the development stages such as investments costs, concept/ strategy development as well as the current investor income and property value. The subjects for this analysis are a shopping centre designed with passive cooling/bioclimatic strategies evaluated at the decision making stage, a certified regional shopping centre and a non-certified standard regional shopping centre. Moreover, the proposal intends to provide decision makers with some tools for linking green design features to the investment analysis in order to optimize the decision making process when looking into cost savings and design quality.

Extended phone log-likelihood ratio features and acoustic-based I-vectors for language recognition

This paper presents new techniques with relevant improvements added to the primary system presented by our group to the Albayzin 2012 LRE competition, where the use of any additional corpora for training or optimizing the models was forbidden. In this work, we present the incorporation of an additional phonotactic subsystem based on the use of phone log-likelihood ratio features (PLLR) extracted from different phonotactic recognizers that contributes to improve the accuracy of the system in a 21.4% in terms of Cavg (we also present results for the official metric during the evaluation, Fact). We will present how using these features at the phone state level provides significant improvements, when used together with dimensionality reduction techniques, especially PCA. We have also experimented with applying alternative SDC-like configurations on these PLLR features with additional improvements. Also, we will describe some modifications to the MFCC-based acoustic i-vector system which have also contributed to additional improvements. The final fused system outperformed the baseline in 27.4% in Cavg.

Learning 3D structure from 2D images using LBP features

An automatic machine learning strategy for computing the 3D structure of monocular images from a single image query using Local Binary Patterns is presented. The 3D structure is inferred through a training set composed by a repository of color and depth images, assuming that images with similar structure present similar depth maps. Local Binary Patterns are used to characterize the structure of the color images. The depth maps of those color images with a similar structure to the query image are adaptively combined and filtered to estimate the final depth map. Using public databases, promising results have been obtained outperforming other state-of-the-art algorithms and with a computational cost similar to the most efficient 2D-to-3D algorithms.

Modeling of GRBAS perceptual evaluation using spectral features obtained from an auditory-based filterbank.

Perceptual voice evaluation according to the GRBAS scale is modelled using a linear combination of acoustic parameters calculated after a filter-bank analysis of the recorded voice signals. Modelling results indicate that for breathiness and asthenia more than 55% of the variance of perceptual rates can be explained by such a model, with only 4 latent variables. Moreover, the greatest part of the explained variance can be attributed to only one or two latent variables similarly weighted by all 5 listeners involved in the experiment. Correlation factors between actual rates and model predictions around 0.6 are obtained.

Definition of masks related to psychovisual features for video quality assessment

Video Quality Assessment needs to correspond to human perception. Pixel-based metrics (PSNR or MSE) fail in many circumstances for not taking into account the spatio-temporal property of human's visual perception. In this paper we propose a new pixel-weighted method to improve video quality metrics for artifacts evaluation. The method applies a psychovisual model based on motion, level of detail, pixel location and the appearance of human faces, which approximate the quality to the human eye's response. Subjective tests were developed to adjust the psychovisual model for demonstrating the noticeable improvement of an algorithm when weighting the pixels according to the factors analyzed instead of treating them equally. The analysis developed demonstrates the necessity of models adapted to the specific visualization of contents and the model presents an advance in quality to be applied over sequences when a determined artifact is analyzed.

Unique molecular surface features of in vivo tolerized T cells

Differential expression of surface markers can frequently be used to distinguish functional subsets of T cells, yet a surface phenotype unique to T cells induced into an anergic state has not been described. Here, we report that CD4 T cells rendered anergic in vivo by superantigen can be identified by loss of the 6C10 T cell marker. Inoculation of Vβ8.1 T cell antigen receptor (TCR) transgenic mice with a Vβ8.1-reactive minor lymphocyte-stimulating superantigen (Mls-1a) induces tolerance to Mls-1a by clonal anergy. CD4 lymph node T cells from Mls-1a inoculated transgenic mice enriched for the 6C10− phenotype neither proliferate nor produce interleukin-2 upon TCR engagement, whereas 6C10+ CD4 T cells retain responsiveness. Analysis of T cell memory markers demonstrate that 6C10− T cells remain 3G11hi but express heterogeneous levels of CD45RB, CD62L, CD44, and the CD69 early activation marker, suggesting that T cells at various degrees of activation can be functionally anergic. These studies demonstrate that anergic T cells can be purified based on 6C10 expression permitting examination of issues concerning biochemical and biological features specific to T cell anergy.

The specific features of methionine biosynthesis and metabolism in plants

Plants, unlike other higher eukaryotes, possess all the necessary enzymatic equipment for de novo synthesis of methionine, an amino acid that supports additional roles than simply serving as a building block for protein synthesis. This is because methionine is the immediate precursor of S-adenosylmethionine (AdoMet), which plays numerous roles of being the major methyl-group donor in transmethylation reactions and an intermediate in the biosynthesis of polyamines and of the phytohormone ethylene. In addition, AdoMet has regulatory function in plants behaving as an allosteric activator of threonine synthase. Among the AdoMet-dependent reactions occurring in plants, methylation of cytosine residues in DNA has raised recent interest because impediment of this function alters plant morphology and induces homeotic alterations in flower organs. Also, AdoMet metabolism seems somehow implicated in plant growth via an as yet fully understood link with plant-growth hormones such as cytokinins and auxin and in plant pathogen interactions. Because of this central role in cellular metabolism, a precise knowledge of the biosynthetic pathways that are responsible for homeostatic regulation of methionine and AdoMet in plants has practical implications, particularly in herbicide design.

Quantifying pursuit-diving seabirds’ associations with fine-scale physical features in tidal streamenvironments

Acknowledgements James J. Waggitt was funded by a NERC Case studentship supported by OpenHydro Ltd and Marine Scotland Science (NE/J500148/1). Vessel-based transects were funded by a NERC (NE/J004340/1) and a Scottish National Heritage (SNH) grant. FVCOM modelling was funded by a NERC grant (NE/J004316/1). Marine Scotland Science provided time on the FRV Alba-na-Mara as part as the Marine Collaboration Research Forum (MarCRF). The bathymetry data used in hydrodynamic models (HI 1122 Sanday Sound to Westray Firth) was collected by the Maritime & Coastguard Agency (MCA) as part of the UK Civil Hydrography Programme. We wish to thank Christina Bristow, Matthew Finn and Jennifer Norris at the European Marine Energy Centre (EMEC); Marianna Chimienti, Ciaran Cronin, Tim Sykes and Stuart Thomas for performing vessel-based transects; Marine Scotland Science staff Eric Armstrong, Ian Davies, Mike Robertson, Robert Watret and Michael Stewart for their assistance; Shaun Fraser, Pauline Goulet, Alex Robbins, Helen Wade and Jared Wilson for invaluable discussions; Thomas Cornulier, Alex Douglas, James Grecian and Samantha Patrick for their help with statistical analysis; and Gavin Siriwardena, Leigh Torres, Mark Whittingham and Russell Wynn for their constructive comments on earlier versions of this manuscript.

Structural features of the γ subunit of the Escherichia coli F1 ATPase revealed by a 4.4-Å resolution map obtained by x-ray crystallography

The F1 part of the F1FO ATP synthase from Escherichia coli has been crystallized and its structure determined to 4.4-Å resolution by using molecular replacement based on the structure of the beef-heart mitochondrial enzyme. The bacterial F1 consists of five subunits with stoichiometry α3, β3, γ, δ, and ɛ. δ was removed before crystallization. In agreement with the structure of the beef-heart mitochondrial enzyme, although not that from rat liver, the present study suggests that the α and β subunits are arranged in a hexagonal barrel but depart from exact 3-fold symmetry. In the structures of both beef heart and rat-liver mitochondrial F1, less than half of the structure of the γ subunit was seen because of presumed disorder in the crystals. The present electron-density map includes a number of rod-shaped features which appear to correspond to additional α-helical regions within the γ subunit. These suggest that the γ subunit traverses the full length of the stalk that links the F1 and FO parts and makes significant contacts with the c subunit ring of FO.

Acute leukemia with promyelocytic features in PML/RARα transgenic mice

Acute promyelocytic leukemia (APL) is associated with reciprocal chromosomal translocations involving the retinoic acid receptor α (RARα) locus on chromosome 17. In the majority of cases, RARα translocates and fuses with the promyelocytic leukemia (PML) gene located on chromosome 15. The resulting fusion genes encode the two structurally unique PML/RARα and RARα/PML fusion proteins as well as aberrant PML gene products, the respective pathogenetic roles of which have not been elucidated. We have generated transgenic mice in which the PML/RARα fusion protein is specifically expressed in the myeloid-promyelocytic lineage. During their first year of life, all the PML/RARα transgenic mice have an abnormal hematopoiesis that can best be described as a myeloproliferative disorder. Between 12 and 14 months of age, 10% of them develop a form of acute leukemia with a differentiation block at the promyelocytic stage that closely mimics human APL even in its response to retinoic acid. Our results are conclusive in vivo evidence that PML/RARα plays a crucial role in the pathogenesis of APL.

Activation of a heterologously expressed octopamine receptor coupled only to adenylyl cyclase produces all the features of presynaptic facilitation in Aplysia sensory neurons

Short-term behavioral sensitization of the gill-withdrawal reflex after tail stimuli in Aplysia leads to an enhancement of the connections between sensory and motor neurons of this reflex. Both behavioral sensitization and enhancement of the connection between sensory and motor neurons are importantly mediated by serotonin. Serotonin activates two types of receptors in the sensory neurons, one of which is coupled to the cAMP/protein kinase A (PKA) pathway and the other to the inositol triphosphate/protein kinase C (PKC) pathway. Here we describe a genetic approach to assessing the isolated contribution of the PKA pathway to short-term facilitation. We have cloned from Aplysia an octopamine receptor gene, Ap oa1, that couples selectively to the cAMP/PKA pathway. We have ectopically expressed this receptor in Aplysia sensory neurons of the pleural ganglia, where it is not normally expressed. Activation of this receptor by octopamine stimulates all four presynaptic events involved in short-term synaptic facilitation that are normally produced by serotonin: (i) membrane depolarization; (ii) increased membrane excitability; (iii) increased spike duration; and (iv) presynaptic facilitation. These results indicate that the cAMP/PKA pathway alone is sufficient to produce all the features of presynaptic facilitation.

Pronounced and sustained central hypernoradrenergic function in major depression with melancholic features: Relation to hypercortisolism and corticotropin-releasing hormone

Both stress-system activation and melancholic depression are characterized by fear, constricted affect, stereotyped thinking, and similar changes in autonomic and neuroendocrine function. Because norepinephrine (NE) and corticotropin-releasing hormone (CRH) can produce these physiological and behavioral changes, we measured the cerebrospinal fluid (CSF) levels each hour for 30 consecutive hours in controls and in patients with melancholic depression. Plasma adrenocorticotropic hormone (ACTH) and cortisol levels were obtained every 30 min. Depressed patients had significantly higher CSF NE and plasma cortisol levels that were increased around the clock. Diurnal variations in CSF NE and plasma cortisol levels were virtually superimposable and positively correlated with each other in both patients and controls. Despite their hypercortisolism, depressed patients had normal levels of plasma ACTH and CSF CRH. However, plasma ACTH and CSF CRH levels in depressed patients were inappropriately high, considering the degree of their hypercortisolism. In contrast to the significant negative correlation between plasma cortisol and CSF CRH levels seen in controls, patients with depression showed no statistical relationship between these parameters. These data indicate that persistent stress-system dysfunction in melancholic depression is independent of the conscious stress of the disorder. These data also suggest mutually reinforcing bidirectional links between a central hypernoradrenergic state and the hyperfunctioning of specific central CRH pathways that each are driven and sustained by hypercortisolism. We postulate that α-noradrenergic blockade, CRH antagonists, and treatment with antiglucocorticoids may act at different loci, alone or in combination, in the treatment of major depression with melancholic features.

Conserved structural features of the synaptic fusion complex: SNARE proteins reclassified as Q- and R-SNAREs

SNARE [soluble NSF (N-ethylmaleimide-sensitive fusion protein) attachment protein receptor] proteins are essential for membrane fusion and are conserved from yeast to humans. Sequence alignments of the most conserved regions were mapped onto the recently solved crystal structure of the heterotrimeric synaptic fusion complex. The association of the four α-helices in the synaptic fusion complex structure produces highly conserved layers of interacting amino acid side chains in the center of the four-helix bundle. Mutations in these layers reduce complex stability and cause defects in membrane traffic even in distantly related SNAREs. When syntaxin-4 is modeled into the synaptic fusion complex as a replacement of syntaxin-1A, no major steric clashes arise and the most variable amino acids localize to the outer surface of the complex. We conclude that the main structural features of the neuronal complex are highly conserved during evolution. On the basis of these features we have reclassified SNARE proteins into Q-SNAREs and R-SNAREs, and we propose that fusion-competent SNARE complexes generally consist of four-helix bundles composed of three Q-SNAREs and one R-SNARE.