|
|
Reverse engineering of an audio mix using neural networks
Čermák, Jerguš ; Mokrý, Ondřej (referee) ; Rajmic, Pavel (advisor)
This bachelor's thesis focuses on the use of neural network algorithms to determine the parameters of signal processors used in the mixing of audio tracks. The thesis presents linear signal processors such as \emph{Gain}, \emph{Pan}, \emph{Filter}, and \emph{Reverb}, which are commonly used to process audio signal and to produce a stereo mix of the audio recording. These processors are subsequently used within the neural network model as layers, implemented using the DDSP library, aimed at predicting the parameters used in the mix, given the knowledge of the input tracks and the target mix. Resultantly, stereo mixdowns were created, and their parameters were estimated using two neural network models. The results were evaluated using both objective measurements and subjective methods (listening test).
|
|
| |
|
|
Reconstruction of corrupted audio signals using deep unrolling
Kment, František ; Myška, Vojtěch (referee) ; Mokrý, Ondřej (advisor)
The thesis deals with the problem of audio signal restoration using traditional optimization methods combined with deep unrolling methods. An optimization task for filling in missing sections of the audio signal was formulated, and the proximal algorithm FISTA was chosen and subsequently implemented. Furthermore, three unrolled variants of the algorithm (Unrolled Fista Net) were implemented, two of which were further optimized using tests on a selected test dataset and trained on the Nsynth dataset. The results of the trained networks were compared both with competing methods and the original untrained variant of the algorithm. The comparison was made using objective metrics (MSE, SNR, PEAQ, PEMO-Q) and a subjective listening test.
|
|
|
Methods of decomposition of musical signals into harmonic and transient components
Guľas, Matej ; Rajmic, Pavel (referee) ; Mokrý, Ondřej (advisor)
The topic of this thesis is to investigate the separation of an audio signal into harmonic and transient components, and will focus on two separation methods, namely separation using a median filter and separation using phase-corrected convex optimization. In the first part, the theoretical information and concepts necessary to handle this task are covered and in the second part, the given methods and their variants are implemented and also a comparative study using subjective listening and objective metrics is elaborated.
|
|
|
Audio enhancement using different-quality A/D converters
Kovanda, Vojtěch ; Mokrý, Ondřej (referee) ; Rajmic, Pavel (advisor)
This paper deals with sparsity based audio dequantization using combination of few cheaper and one more expensive A/D converters. Paper finds solution of this problem as convex optimization using generic proximal algorithm implemented in MATLAB. Real and synthetic signal reconstructions were performed using this algorithm. The reconstruction performance is evaluated using objective criteria such as Signal-to-Distortion Ratio (SDR) and PEMO-Q, which consider the perceptual properties of the signals.
|
|
|
Modern Optimization Methods for Interpolation of Missing Sections in Audio Signals
Mokrý, Ondřej ; Kowalski, Matthieu (referee) ; Koldovský, Zbyněk (referee) ; Rajmic, Pavel (advisor)
Poškození audio signálů je v praxi běžným, avšak nežádoucím faktem. Ke ztrátě informace může dojít nevhodným záznamem (nízký vzorkovací kmitočet či dynamický rozsah), chybou přenosu (výpadek vzorků), poškozením média či z důvodu rušení. Odstraňování takových poruch je možné pomocí inverzních úloh. Tato práce se konkrétně zaměřuje na situaci, kdy jsou úseky audio signálu o délce v řádu desítek milisekund zcela ztraceny a cílem je chybějící vzorky interpolovat na základě kontextu a vhodného modelu signálu. První část dizertační práce se věnuje metodám konvexní i nekonvexní optimalizace, které hledají řešení interpolační úlohy na základě předpokladu řídkosti časově-kmitočtového spektra. Obecný základ i některé algoritmy jsou převzaté z literatury a přizpůsobené interpolační úloze, řada modifikací a experimentálních přístupů je originální. Druhá část práce je zaměřena na využití nezáporné faktorizace matic, s níž lze sestavit pravděpodobnostní model spektrogramu signálu a tento využít pro jeho interpolaci. Z tohoto modelu pak vychází úspěšný rekonstrukční algoritmus, k němuž jsou v této práci odvozeny dvě alternativní metody. Závěr práce se věnuje rozsáhlému experimentálnímu ověření funkčnosti metod na skupině hudebních signálů. S využitím objektivních ukazatelů kvality interpolovaného signálu je ukázáno, že v jednotlivých třídách metod vedou navržené modifikace ke znatelnému zlepšení kvality či zlepšení konvergence oproti metodám základním. V rámci studovaného rozsahu poškození pak zejména algoritmy využívající faktorizace konkurují současným nejlepším metodám pro interpolaci chybějících úseků audio signálu.
|
|
|
Web applications supporting education of audio signal processing
Malcher, Tomáš ; Mokrý, Ondřej (referee) ; Rajmic, Pavel (advisor)
This bachelors thesis deals with a creation of web applications using JavaScript language. These web applications should be helpful in signal processing classes by showing and explaining different phenomenons to students. Students can see how different parameters affect the outcome of said issue. The theory needed for understanding and implementation is covered by this thesis. Thesis covers key parts of the code of said web applications and the libraries used for their implementaition.
|
|
|
Restoration of degraded audiosignals using sparse representations
Mokrý, Ondřej ; Veselý, Vítězslav (referee) ; Rajmic, Pavel (advisor)
This bachelor thesis is focused on the problem of inpainting a segment of missing samples in an audiosignal. The signal is represented as sparse vector using discrete Gabor transform. The problem of inpainting missing samples while preserving the sparsity of the representation is formulated as an optimisation task, which is then solved using Douglas-Rachford algorithm. In contrast with the state-of-the-art approaches, the algorithm is extended by proposing method for compensating the energy decrease which occurs in the restored signal.
|
|
|
Modern methods of reconstruction of saturated audio signals
Mikulášková, Aneta ; Mokrý, Ondřej (referee) ; Rajmic, Pavel (advisor)
This thesis presents a variety of soft clipping simulations and their respective de-clipping functions. A restoration method uses a Douglas-Rachford algorithm (already verified for hard de-clip). The algorithm is extended by inverse functions and simultaneous de-quantization is proposed. The simulations are applied to artificial signals, and audio samples, and the proposal is tested in the computing environment MATLAB. A restored signal is later evaluated by objective and subjective methods.
|
|
|
Interactive web applications supporting education of 3D graphics
Morávek, Jan ; Mokrý, Ondřej (referee) ; Rajmic, Pavel (advisor)
This diploma thesis is focused on computer 3D graphics and the implementation of educational applications in JavaScript language. Discussed topics of computer graphics include object transformations, Bezier patches and the role of the camera in the scene. The thesis describes the basic theory of these areas and educational applications. The thesis also includes the detailed description of the features and the implementation of the created applications. In the end of the thesis possible extensions are discussed.
|
guest ::
RSS feed.