|
|
Machine learning in audio effects
Sychra, Jakub ; Mošner, Ladislav (oponent) ; Černocký, Jan (vedoucí práce)
Reverse engineering audio effects from mixed tracks is a complex topic requiring signal processing and music engineering experience. This work aims at creation of a system capable of identifying the sequence and parameters of guitar effects from a mixed audio track. Training data was created using clean guitar sounds from IDMT-SMT-Audio-Effects, augmented by known effects (BitCrush, Chorus, Clipping, Compressor, Delay, Distortion, High-pass filter, Ladder filter, Low-pass filter, Limiter, Phaser and Reverb), all implemented with a Python wrapper around standard VST effects. The system is based on VGGish neural network architecture with several classification (presence of effects) and regression (parameters of effects) heads. The performance of the algorithm is evaluated on classification and regression accuracy, as well as in informal listening tests.
|
|
|
Differential-based deepfake speech detection
Staněk, Vojtěch ; Černocký, Jan (oponent) ; Firc, Anton (vedoucí práce)
Deepfake speech technology, which can create highly realistic fake audio, poses significant challenges, from enabling multi-million dollar scams to complicating legal evidence's reliability. This work introduces a novel method for detecting such deepfakes by leveraging bonafide speech samples. Unlike previous strategies, the approach uses trusted ground truth speech samples to identify spoofs, providing critical information that common methods lack. By comparing the bonafide samples with potentially manipulated ones, the aim is to effectively and reliably determine the authenticity of the speech. Results suggest that this innovative approach could be a valuable tool in identifying deepfake speech, especially recordings created using Voice Conversion techniques, offering a new line of defence against this emerging threat.
|
|
|
Whispered to Normal Speech Conversion
Gajda, Richard ; Černocký, Jan (oponent) ; Brukner, Jan (vedoucí práce)
The goal of this thesis is to develop an alternative solution to parallel datasets needed for whispered to normal voice speech conversion, using pseudo-whispered speech synthesis. Psuedo-whispered datasets generated from open source voice datasets are used to train a voice conversion model (vocoder BigVGAN) which is both speaker and language independent. Resulting model is compared to a baseline and existing solutions.
|
|
|
Very Low Bit-Rate Speech Coding Based on Neural Networks
Jochman, Stanislav ; Malenovský, Vladimír (oponent) ; Černocký, Jan (vedoucí práce)
During this work, we focused on replicating and enhancing results by using the neural network LPCNet. We compared audio quality from the pre-trained model and our models trained on smaller datasets, thus reducing training time and improving audio quality. We determined that using a language-specific dataset can produce greater results in that specific language than a big general model. We measured the quality of speech of the pre-trained model and our models using WARPQ ranking score 5.2.4. We also examined possibilities of improving audio quality by filtering output audio using output post-filters and formant-enhancing filters. Our results show measurable improvement in audio quality using the suggested methods.
|
|
|
Automatická tvorba varhanní předehry k církevním písním
Maňák, Ondřej ; Černocký, Jan (oponent) ; Fapšo, Michal (vedoucí práce)
Tato diplomová práce se zabývá problematikou automatické tvorby varhanních předeher k církevním písním z teoretického i praktického hlediska. Varhanní předehra je krátký úvod k liturgickému zpěvu. Vzhledem k tomu, že ji lze popsat pomocí konečné množiny pravidel, nabízí se k její tvorbě využít principů úloh s omezujícími podmínkami. Efektivním prostředkem pro vývoj systému potom může být programovací jazyk C++ a knihovna Gecode.
|
|
|
Rozpoznávaní tónů a akordů na platformě Android
Kucharovič, Oliver ; Černocký, Jan (oponent) ; Szőke, Igor (vedoucí práce)
Cieľom tejto práce je rozpoznávanie tónov a akordov na ľubovoľnom hudobnom nástroji pomocou mikrofónu v mobilných zariadeniach Android. Aplikácia sníma alebo nahráva zvuky, ktoré analyzuje a zapisuje do zápisu podobnej hudobnej notovej osnove v reálnom čase. Práca popisuje problematiku rozpoznávania a podkladá riešenie danej problematiky pomocou rýchlej Fourierovej transformácie a analýzy spektra hlasitostí. Ďalej práca popisuje hudobnú teóriu a trénovanie jednotlivých tónov. Aplikácia bola podrobená užívateľským testom a testom na presnosť rozpoznávania. Záver obsahuje výsledky z testovania a štatistiky z publikácie na Google Play.
|
|
| |
|
|
Improving Robustness of Speaker Recognition using Discriminative Techniques
Novotný, Ondřej ; Ferrer, Luciana (oponent) ; Pollák, Petr (oponent) ; Černocký, Jan (vedoucí práce)
This work deals with discriminative techniques in speaker verification systems to improve robustness of the systems against factors that negatively affect their performance. These factors include noise, reverberation, or the transmission channel. The thesis consists of two main parts. In the first part, it deals with a theoretical introduction to current state-of-the-art speaker verification systems. The recognition system's steps are described, starting from the extraction of acoustic features, the extraction of vector representations of recordings, and the final recognition score computation. Particular emphasis is paid to the techniques of extraction of a vector representation of a recording, where we describe two different paradigms: the i-vectors and the x-vectors. The second part of the work focuses more on discriminative techniques to increase robustness. Their description is organized to match the gradual passage of the recording through the verification system. First, attention is paid to signal pre-processing using a neural network for noise reduction and speech enhancement. This pre-processing is a universal technique independent of the verification system. The work follows by focusing on the use of a discriminative approach in the extraction of features and the extraction of vector representations of recordings. Furthermore, this work sheds light on the transition from generative systems to discriminative systems. In order to give a fuller context, the work also describes techniques that had historically preceded this transition. All presented techniques are always experimentally verified and their advantages evaluated. We are proposing several techniques that have proved successful in both the generative approach in the form of i-vectors and discriminative x-vectors, and thanks to them, considerable improvement has been achieved. For completeness, in the field of robustness, other techniques are included in the work, such as normalization of scores or multi-condition training. Finally, the work deals with the robustness of discriminative systems in terms of data used in their training.
|
|
|
Chatbot Capable of Information Search
Ďurista, Michal ; Beneš, Karel (oponent) ; Černocký, Jan (vedoucí práce)
''Chatbot'' is a very popular term in today's artificial intelligence era. Chatbots can be seen in business solutions more a more nowadays. The main goal of this thesis is to create an algorithm that is capable of information retrieval and implement it into a chatbot. The information resides on a real customer's web pages. The thesis also provides an overview of current chatbot situation along with the Microsoft technologies used for the development. The technological background of these technologies, mostly natural language processing techniques, is covered too. The thesis also describes the implementation of the algorithm and the chatbot itself as well as the real industrial environment testing process.
|
|
|
Automatické hledání vazeb mezi částmi audiovizuálních dokumentů
Sychra, Marek ; Černocký, Jan (oponent) ; Szőke, Igor (vedoucí práce)
Tato práce se zabývá tématem hledání tématu v textu. Konkrétně hledání spojitostí mezi krátkými texty a hledání hranic jednotlivých částí stejného tématu v jednom hlavním textu. Hlavní motivací výzkumu bylo zavedení do praxe a to v rámci aplikace na přednáškové materiály na FIT (provázání jednotlivých částí různých přednášek). Přístup k porovnávání textů spočívá v analýze textu a slov, která obsahuje a zjišťování významu a důležitosti jednotlivých slov. Segmentace textu toto využívá, když hledá předěly mezi tématy v textu. Obě části problému ( link detection, story segmentation ) měly velmi vysokou úspěšnost na testovacích datech (zprávy ze světových novin). Při subjektivním vyhodnocování u částí přednášek byla úspěšnost nižší, ale stále dobrá.
|
host ::
RSS.