Národní úložiště šedé literatury Nalezeno 7 záznamů.  Hledání trvalo 0.01 vteřin. 
Penetration Tests of Speaker Verification System
Wojnar, Filip ; Landini, Federico Nicolás (oponent) ; Plchot, Oldřich (vedoucí práce)
The aim of the thesis is to realize penetration tests of automatic speaker verification system with use of text-to-speech model. The thesis is focused on inner functioning of those systems and spoofing attacks against them. The thesis is also focused on speech synthesis. Later chapters are focused on realization of realized penetration tests and discussion about results they brought us.
Personal Voice Activity Detection
Sedláček, Šimon ; Landini, Federico Nicolás (oponent) ; Švec, Ján (vedoucí práce)
This work aims to implement, test, and evaluate a speaker-conditioned Voice Activity Detection (VAD) method called Personal VAD. The method builds upon an LSTM-based approach to VAD and its purpose is to introduce a system that can reliably detect speech of a target speaker, while retaining the typical characteristics of a VAD system, mainly in terms of small model size, low latency, and low necessary computational resources. The system is trained to distinguish between three classes: non-speech, target speaker speech, and non-target speaker speech. For this purpose, the method utilizes speaker embeddings as a part of the input feature vector to represent the target speaker. Some of the more heavyweight personal VAD variants also make use of speaker verification scores issued to each frame based on the target embedding, resulting in a more robust system. In addition to the one scoring method presented in the original article, two other scoring approaches are introduced, both outperforming the baseline method and improving the performance even for acoustically challenging conditions.
Exploratory Analysis of Big Data in "jmenomesto"
Trampeška, Václav ; Plchot, Oldřich (oponent) ; Landini, Federico Nicolás (vedoucí práce)
This thesis is concerned with the analysis of the database of the online game Jméno, město. In this game, players are tasked with adequately answering given categories with answers beginning with a given letter. The thesis analyzes the evolution of player behaviour over the lifetime of the game and the behaviour of players in different countries and cultures within the same and different languages based on the popularity of different answers. A web application was developed to facilitate the execution of these analyses, allowing easy-to-use data collection and data visualisation in charts without requiring knowledge of the database structure and a query language. The results of the proposed analysis methods are compared with Google Trends data to identify the similarities between the data observed in the game and internet searches. The comparison shows that the proposed methods can give meaningful results, and hence the database can be suitable for performing further specific analyses. Furthermore, partly based on the analysis results, the thesis proposes changes to improve the game in player experience and revenue generation.
Gaussian Processes Based Hyper-Optimization of Neural Networks
Coufal, Martin ; Landini, Federico Nicolás (oponent) ; Beneš, Karel (vedoucí práce)
The goal of this thesis is to create a lightweight toolkit for artificial neural network hyper-parameter optimisation. The optimisation toolkit has to be able to optimise multiple, possibly correlated hyper-parameters. I solved this problem by creating an optimiser that uses Gaussian processes to predict the influence of the hyper-parameters on the resulting neural network accuracy. Based on the experiments on multiple benchmark functions, the toolkit is able to provide better results than random search optimisation and thus reduce the number of necessary optimisation steps. The random search optimisation provided better results only in the first few optimisation steps before Gaussian process optimisation creates sufficient model of the problem. However the experiments on MNIST dataset show that random optimisation achieves almost always better results than used GP optimiser. These differences between the experiments results are probably caused by insufficient complexity of the benchmarks or by selected parameters of the implemented optimiser.
Non-Supervised Sentiment Analysis
Karabelly, Jozef ; Landini, Federico Nicolás (oponent) ; Fajčík, Martin (vedoucí práce)
The goal of this thesis is to present an overview of the current state of research in the non-supervised sentiment analysis and identify potential research paths. Besides, the thesis introduces a novel self-supervised pre-training objective. Extending the model trained with the introduced objective with one extra layer of neural network and training it alone shows promising results.  The extended model indicates an ability to encode the abstract representation of overall sentiment, emotions and sarcasm. A custom dataset was specifically collected for the pre-training objective introduced in this thesis. Future improvements and possible research paths are proposed based on the experiments performed with the extended model.
Machine Comprehension Using Commonsense Knowledge
Daniš, Tomáš ; Landini, Federico Nicolás (oponent) ; Fajčík, Martin (vedoucí práce)
In this thesis, the commonsense reasoning ability of modern neural systems is explored. The goal is to provide insight into the current state of research in this area and identify promising research directions. A state-of-the-art question-answering model has been implemented and experimented with in various scenarios. Unlike in older approaches, the model achieved comparable results with best available models for the target task without using any task-specific architecture. Furthermore, unintended statistical biases are discovered in a popular commonsense reasoning dataset which allow models to compute the correct answer even when it does not have sufficient information to do so. Based on these findings, recommendations and possible future research areas are suggested.
Modelování hudby na úrovni signálu pomocí WaveNetu
Slanináková, Terézia ; Landini, Federico Nicolás (oponent) ; Beneš, Karel (vedoucí práce)
Práca sa zaoberá skúmaním možnosti modelovania hudby a reči pomocou WaveNetu, hlbokou neurónovou sieťou pre generovanie zvuku na úrovni signálu. Za pomoci existujúcich implementácií bol WaveNet netrénovaný na rôznych datasetoch a vyprodukoval mnohé zvukové súbory. Bolo vykonaných niekoľko experimentov s rôznym nastavením hyperparametrov WaveNetu. Taktiež bolo použitých niekoľko schém generovania, každá s rôznym vplyvom na generovaný výsledok. Kvalita výstupných zvukových súborov bola ohodnotená na základe dotazníku. Hudobné zvukové stopy dosiahli skóre 2-3.1818 na 5-bodovej škále, čo je porovnateľné s  hudobnými nahrávkami originálneho výskumného tímu (3.1818).

Chcete být upozorněni, pokud se objeví nové záznamy odpovídající tomuto dotazu?
Přihlásit se k odběru RSS.