1 |
A comparative analysis of gaussian mixture models and i-vector for speaker verification under mismatched conditionsAvila, Anderson Raymundo January 2014 (has links)
Orientador: Prof. Dr. Francisco J. Fraga / Dissertação (mestrado) - Universidade Federal do ABC, Programa de Pós-Graduação em Engenharia da Informação, 2014. / Most speaker verifcation systems are based on Gaussian mixture models and more
recently on the so-called i-vector. These two methods are affected in mismatched testtrain
conditions, which might be caused by vocal-efort variability, different speakingstyles
or channel efects. In this work, we compared the impact of speech rate variation
and room reverberation on both methods. We found that performance degradation
due to variation on speech rate can be mitigated by adding fast speech samples into
the training set, which decreased equal error rates for Gaussian mixture models and
i-vector, respectively. Regarding reverberation, we investigated the achievements of
both methods when three diferent reverberation compensation techniques are applied
in order to overcome performance degradation. The results showed that having
reverberant background models separated by diferent levels of reverberation can bene
t both methods, with the i-vector providing the best performance in that scenario.
Finally, the performance of two auditory-inspired features, mel-frequency cepstral coe
ficients and the so-called modulation spectrum features, are compared in presence
of room reverberation. For the speaker verifcation system considered in this work,
modulation spectrum features are equally afected by reverberation time and have
their performance degraded as the level of reverberation increases.
|
2 |
Testes de design: Uma abordagem baseada em testes para verificação automática de conformidade estrutural entre implementação e regras de design. / Design tests: A test-based approach for automatic verification of structural compliance between implementation and design rules.MONTEIRO, João Arthur Brunet. 22 August 2018 (has links)
Submitted by Johnny Rodrigues (johnnyrodrigues@ufcg.edu.br) on 2018-08-22T23:26:49Z
No. of bitstreams: 1
JOÃO ARTHUR BRUNET MONTEIRO - DISSERTAÇÃO PPGCC 2010..pdf: 1055709 bytes, checksum: e8145387f1111c96339e77c6b84ebade (MD5) / Made available in DSpace on 2018-08-22T23:26:49Z (GMT). No. of bitstreams: 1
JOÃO ARTHUR BRUNET MONTEIRO - DISSERTAÇÃO PPGCC 2010..pdf: 1055709 bytes, checksum: e8145387f1111c96339e77c6b84ebade (MD5)
Previous issue date: 2010-07-26 / Assegurar que um programa está de acordo com sua especificação é um elemento chave na garantia de qualidade de software. Embora haja amplo suporte ferramental para checar se uma implementação está funcionalmente de acordo com seus requisitos, checar se ela está em conformidade com regras de design ainda é uma atividade que por muitas vezes é executada manualmente. Neste trabalho propomos uma técnica que visa automatizar a checagem de conformidade entre regras de design de baixo-nível e implementação. A técnica proposta, intitulada testes de design, permite a verificação de programas baseada em testes. O objetivo da técnica é checar se os programadores estão seguindo as regras de design previamente especificadas. Regras são especificadas como testes, daí o nome testes de design. De fato, testes de design são JUnit test cases com uma semântica diferente de testes funcionais. Testes funcionais checam se o software se comporta como esperado quando estimulado por determinadas entradas, ao passo que testes de design checam se o software está sendo construído da maneira esperada. Para dar suporte à abordagem de testes de design, foi desenvolvida uma biblioteca chamada DesignWizard (http://www.designwizard.org). Uma avaliação levando em consideração a usabilidade e a escalabilidade do DesignWizard foi efetuada com o objetivo de mostrar sua viabilidade na checagem de conformidade de grandes projetos. Os resultados do experimento de escalabilidade apontam na direção de que a eficiência não é um problema para a ferramenta e que, de acordo com o experimento de usabilidade, o suporte da ferramenta à escrita de testes de design cumpre com as expectativas dos desenvolvedores. / Assuring that a program conforms to its specification is a key concern in software quality
assurance. Although there is substantial tool support to check whether an implementation
complies to its functional requirements, checking whether it conforms to its design remains as an almost completely manual activity. We propose the concept of design tests, which are test-like programs that automatically check whether an implementation conforms to a specific design rule. Design rules are implemented directly in the target programming language in the form of tests. As a proof of concept, we present DesignWizard, an library developed to support design tests for Java programs as JUnit test cases. We have performed an evaluation to show that DesignWizard’s API is easy to use and the tool scales as software grows. To achieve this, we have conducted an experiment to assess the usability of DesignWizard’s API by analysing eleven developers on the activity of composing five design tests using DesignWizard’s API. Besides that, we have measured the time and memory consumption that DesignWizard takes to apply static analysis on several projects sizes varying between 0.125MB and 46MB. The results of our study show that the time and memory consumption of static analysis performed by Design Wizard has a linear form as the size of the application grows. Besides that, the experiment conducted with the programmers leads us to conclude that DesignWizard’s API is easy to use in the sense that it meets to programmers expectations.
|
Page generated in 0.3528 seconds