|
| |
|
|
Evaluation of Sources of Human Speech for Deepfake Creation
Frič, Michal ; Malinka, Kamil (oponent) ; Firc, Anton (vedoucí práce)
Voice deepfakes, powered by rapid advancements in artificial intelligence and machine learning, represent a dual-edge technology with significant benefits and risks. These synthetic voice outputs are increasingly realistic due to the easy access to vast amounts of digital speech data from various sources. This thesis analyses these sources’ suitability for creating convincing deepfakes. We identified and evaluated numerous speech sources and developed methodologies for assessing their quality, accessibility, diversity, and update frequency. The evaluation extended to analyzing the impact of source characteristics on deepfake quality and the effectiveness of detection by software and human evaluators. Findings indicate that all identified sources can provide sufficiently high-quality recordings to create high-quality deepfakes, often indistinguishable. Additionally, they highlight each source’s particular strengths and weaknesses (measured properties) grade. An anomaly in detection software was discovered, allowing deepfakes to be modified to evade detection. Furthermore, less than 10 seconds of human speech could suffice to create a high-quality deepfake, directly correlating the length and quality of input recordings to the fidelity of the output. The thesis concludes with a discussion of the risks associated with these sources and proposes measures for prevention and mitigation.
|
|
|
Aplikace posilovaného učení v řízení Smart Home
Biel, Gabriel ; Zbořil, František (oponent) ; Janoušek, Vladimír (vedoucí práce)
This thesis investigates how machine learning can improve smart home management by focusing on optimizing temperature control and boosting energy efficiency. Specifically, it examines and compares two sophisticated reinforcement learning algorithms, Deep Q-Learning (DQL) and Proximal Policy Optimization (PPO). These models are tested in a simulated environment that replicates real-world conditions to evaluate their effectiveness in adapting to user behaviors and environmental changes. The study finds that the PPO model is particularly effective due to its stability and ability to predict when occupants will return, thus maintaining a comfortable temperature more efficiently. This research offers valuable insights into the practical applications of AI technologies in smart homes.
|
|
|
Optimization of DDoS Attack Mitigation based on Machine Learning
Banák, Filip ; Šišmiš, Lukáš (oponent) ; Kučera, Jan (vedoucí práce)
DDoS attacks using the TCP protocol are still amongst the most common. This thesis aims to take advantage of information present in TCP SYN messages to improve DDoS attack detection success rate. TCP SYN fingerprints are proposed as an additional data source to consider when computing features for DDoS detection. A combination of an existing feature extraction and aggregation system with an existing autoencoder-based anomaly detector is significantly optimized and extended to make use of SYN fingerprints. The experimental results show decent DDoS detection improvements on relevant datasets. The detector is 16 and 95 times faster to train and execute respectively. The extraction and aggregation system is 23 times faster.
|
|
|
Implementing gesture recognition on ARM as an alternative to traditional device control
Gajdošík, Richard ; Zbořil, František (oponent) ; Kočí, Radek (vedoucí práce)
This bachelor's thesis focuses on the development and implementation of a gesture recognition system on ARM architecture, utilizing the i.MX 93 board and TensorFlow Lite. The project is grounded in the application of neural networks for the recognition of hand gestures, offering an alternative to traditional device control methods. An integral part of the work involves a comprehensive analysis of existing gesture recognition solutions, identifying their strengths and potential improvements. The thesis elaborates on the design, development, and optimization of a real-time gesture recognition model specifically for ARM chips, emphasizing efficiency and performance. Additionally, the thesis covers the creation of a demonstrative application that visually represents recognized gestures. User testing is conducted to evaluate the practicality and user experience of the gesture recognition system, providing valuable feedback for future enhancements.
|
|
|
Diagnosing anxiety and depression from brain electroencephalogram (EEG) signals
Osvald, Martin ; Jaroš, Marta (oponent) ; Malik, Aamir Saeed (vedoucí práce)
Mental disorders represent inevitable emotions in our society. These psychological states affect the cognitive, emotional and behavioural functioning of individuals. Common men- tal disorders fall into two main diagnostic categories: depressive disorders and anxiety disorders. The aim of this work is to find a new method for detecting whether a given patient suffers from anxiety or depression using EEG classification. In this work, we use a combination of genetic algorithms and models from deep learning.
|
|
|
Creating Novel Deepfake Speech Dataset
Sztolarik, Maroš ; Homoliak, Ivan (oponent) ; Firc, Anton (vedoucí práce)
In the recent years, deepfake technology has advanced to a point where it can convincingly mimic human speech, posing significant challenges in distinguishing between real and synthetic voices. In this thesis, we introduce a novel dataset comprising speech deepfakes generated using diffusion models. This dataset, created with two sophisticated text-to-speech tools, DiffSpeech and ProDiff, aims to provide insight into the threat that these new tools pose. Two more datasets are created with more mature tools, Glow-TTS and Tacotron2, to provide a point of comparison. Then all the generated samples are analyzed through two deepfake detectors in order to provide a direct comparison into how much of a threat each tool is to these detectors. The results show that even though the tools utilizing the diffusion models are threatening, the use of diffusion models did not provide these tools any meaningful advantage in evading the detection.
|
|
|
Creating a Python-based Automated System for Recognizing Emotions from Facial Expressions.
Zima, Samuel ; Malik, Aamir Saeed (oponent) ; Hussain, Yasir (vedoucí práce)
This thesis examines facial expression recognition (FER) using deep learning by focusing on its application in devices with limited memory and computational resources. It begins by researching emotions and facial expressions from psychological, biological, and sociological perspectives. The core of this thesis involves the design and implementation of an automated FER system using the FER-2013 dataset. This system uses a customized SqueezeNet architecture enhanced with a simple bypass, dropout layers and batch normalization layers. This system achieves an accuracy of 66.37 % on the FER-2013 dataset. For comparative analysis, this model was compared with a customized VGG16 architecture which achieved an accuracy of 65.09 %. This thesis provides valuable insights into the development of smaller, more efficient machine learning models for FER which are usable in a wide range of devices, including low-performance CPUs and embedded devices.
|
|
| |
|
|
The use of deep neural networks for the evaluation of metallographic cross-sections
Semančík, Adam ; Mendřický, Radomír (oponent) ; Hurník, Jakub (vedoucí práce)
This thesis explores the application of deep neural networks to improve the evaluation of metallographic cross-sections in materials produced through powder bed fusion. It focuses on two advanced image processing techniques: semantic segmentation and image super-resolution. A U-Net architecture was used for semantic segmentation to classify defects such as lack of fusion porosity and gas porosity. Additionally, an SRGAN (Super-Resolution Generative Adversarial Network) model was utilized to upscale image resolution, potentially enhancing segmentation accuracy. The research assesses whether a model trained on AlSi10Mg can generalize to Cu99 and Ti6Al4V and evaluates the influence of super-resolution on segmentation performance. Results showed that while the segmentation model performed well on AlSi10Mg, generalization to other materials required more diverse training data. Due to computational limitations, the combined effect of super-resolution and segmentation remains inconclusive, suggesting further research with enhanced computational resources.
|
host ::
RSS.