|
|
Speaker Recognition Based on Long Temporal Context
Fér, Radek ; Matějka, Pavel (referee) ; Černocký, Jan (advisor)
Tato práce se zabývá extrakcí vhodných příznaků pro rozpoznávání řečníka z delších časových úseků. Po představení současných technik pro extrakci takových příznaků navrhujeme a popisujeme novou metodu pracující v časovém rozsahu fonémů a využívající známou techniku i-vektorů. Velké úsilí bylo vynaloženo na nalezení vhodné reprezentace temporálních příznaků, díky kterým by mohly být systémy pro rozpoznávání řečníka robustnější, zejména modelování prosodie. Náš přístup nemodeluje explicitně žádné specifické temporální parametry řeči, namísto toho používá kookurenci řečových rámců jako zdroj temporálních příznaků. Tuto techniku testujeme a analyzujeme na řečové databázi NIST SRE 2008. Z výsledků bohužel vyplývá, že pro rozpoznávání řečníka tato technika nepřináší očekávané zlepšení. Tento fakt diskutujeme a analyzujeme ke konci práce.
|
|
|
Assessing movement of articulatory organs based on acoustic analysis of speech
Novotný, Kryštof ; Galáž, Zoltán (referee) ; Mekyska, Jiří (advisor)
Hypokinetic dysarthria is a motor speech disorder often present during Parkinson’s disease. It affects the speech system, including articulatory abilities. There are several speech parameters describing this domain, so it is suggested to deal with their mutual comparison. This work aims to design and describe an algorithm for calculating the parameters of articulation, adapted for the Czech language, and then compare their discriminative power. The acoustic analysis of speech included in it is done via the Praat program and basic machine learning algorithms such as Expectation-Maximization, Kmeans and linear regression are used for the subsequent data processing. The Mann-Whitney U test and representatives of linear, nonlinear and ensemble machine learning models using cross-validation and balanced accuracy are used for evaluation. The results are scripts for automatic assessment of vowel space area, for calculating articulation parameters and for their evaluation. The outputs of the analysis of two different databases (PARCZ and CoBeN) prove that differences in articulation can indeed be observed between normal and dysarthric speech. Based on the mutual comparison of results, it is therefore proposed in the work which parameters and models of machine learning are being appropriate for further dealing with this issue.
|
|
|
Analysis of speech disorders in patients with a high risk of developing Lewy body diseases
Novotný, Kryštof ; Kováč, Daniel (referee) ; Mekyska, Jiří (advisor)
Lewy bodies diseases (one of the most common neurodegenerative disorders) have the same pathological basis, but the individual representatives differ in their clinical manifestations. Different diseases affect the mental or physical side of the patient to a greater or lesser extent. This work assumes that thanks to the acoustic analysis of speech, it is possible to distinguish individual diseases from one another, because the disorders of the cognitive and motor aspects of a patient reflect in speech in specific ways. The thesis aims to describe the clinical features of the main representatives of the Lewy bodies diseases, to investigate their impact on speech, to propose characterizing acoustic parameters and then to compare their discriminative power. Speech recordings from the CoBeN and preLBD databases are used as input data for the proposed algorithm. Descriptive statistics, Mann-Whitney U test, FDR correction and XGBoost machine learning model using stratified cross-validation and balanced accuracy are used for subsequent evaluation. The result are scripts for the automated calculation of speech parameters from the database and their evaluation. The results of the analysis prove that the selected diseases can really be distinguished from each other and from a healthy control based on the manifestations in speech, already in the prodromal stages.
|
|
|
Assessing Movement of Articulatory Organs in Patients with Parkinson’s Disease
Novotný, K. ; Mekyska, J.
Hypokinetic dysarthria is a motor speech disorder often present during Parkinson’s disease. It affects the speech system, including articulatory abilities. There are several speech parameters describing this domain, so it is suggested to deal with their mutual comparison. This work aims to design and describe an algorithm for calculating the parameters of articulation, adapted for the Czech language, and then compare their discriminative power. The acoustic analysis of speech included in it is done via the Praat program and basic machine learning algorithms such as Expectation-Maximization, K-means and linear regression are used for the subsequent data processing. The Mann-Whitney U test, descriptive statistics and Random Forest machine learning model using cross-validation and balanced accuracy is used for evaluation. The results are scripts for automatic assessment of vowel space area, for calculating articulation parameters and for their evaluation. The outputs of the analysis of speech recording database prove that differences in articulation can indeed be observed between normal and dysarthric speech. Based on the mutual comparison of results, it is therefore proposed in the work which parameters are being appropriate for further dealing with this issue.
|
|
|
Assessing movement of articulatory organs based on acoustic analysis of speech
Novotný, Kryštof ; Galáž, Zoltán (referee) ; Mekyska, Jiří (advisor)
Hypokinetic dysarthria is a motor speech disorder often present during Parkinson’s disease. It affects the speech system, including articulatory abilities. There are several speech parameters describing this domain, so it is suggested to deal with their mutual comparison. This work aims to design and describe an algorithm for calculating the parameters of articulation, adapted for the Czech language, and then compare their discriminative power. The acoustic analysis of speech included in it is done via the Praat program and basic machine learning algorithms such as Expectation-Maximization, Kmeans and linear regression are used for the subsequent data processing. The Mann-Whitney U test and representatives of linear, nonlinear and ensemble machine learning models using cross-validation and balanced accuracy are used for evaluation. The results are scripts for automatic assessment of vowel space area, for calculating articulation parameters and for their evaluation. The outputs of the analysis of two different databases (PARCZ and CoBeN) prove that differences in articulation can indeed be observed between normal and dysarthric speech. Based on the mutual comparison of results, it is therefore proposed in the work which parameters and models of machine learning are being appropriate for further dealing with this issue.
|
|
|
Speaker Recognition Based on Long Temporal Context
Fér, Radek ; Matějka, Pavel (referee) ; Černocký, Jan (advisor)
Tato práce se zabývá extrakcí vhodných příznaků pro rozpoznávání řečníka z delších časových úseků. Po představení současných technik pro extrakci takových příznaků navrhujeme a popisujeme novou metodu pracující v časovém rozsahu fonémů a využívající známou techniku i-vektorů. Velké úsilí bylo vynaloženo na nalezení vhodné reprezentace temporálních příznaků, díky kterým by mohly být systémy pro rozpoznávání řečníka robustnější, zejména modelování prosodie. Náš přístup nemodeluje explicitně žádné specifické temporální parametry řeči, namísto toho používá kookurenci řečových rámců jako zdroj temporálních příznaků. Tuto techniku testujeme a analyzujeme na řečové databázi NIST SRE 2008. Z výsledků bohužel vyplývá, že pro rozpoznávání řečníka tato technika nepřináší očekávané zlepšení. Tento fakt diskutujeme a analyzujeme ke konci práce.
|
guest ::
