Among the diverse research fields within computer music, synthesis and generation of audio signals epitomize the cross-disciplinarity of this domain, jointly nourishing both scientific and artistic practices since its creation. Inherent in computer music since its genesis, audio generation has inspired numerous approaches, evolving both with musical practices and scientific/technical advances. Moreover, some syn- thesis processes also naturally handle the reverse process, named analysis, such that synthesis parameters can also be partially or totally extracted from actual sounds, and providing an alternative representation of the analyzed audio signals. On top of that, the recent rise of machine learning algorithms earnestly questioned the field of scientific research, bringing powerful data-centred methods that raised several epistemological questions amongst researchers, in spite of their efficiency. Especially, a family of machine learning methods, called generative models, are focused on the generation of original content using features extracted from an existing dataset. In that case, such methods not only questioned previous approaches in generation, but also the way of integrating this methods into existing creative processes. While these new generative frameworks are progressively introduced in the domain of image generation, the application of such generative techniques in audio synthesis is still marginal. In this work, we aim to propose a new audio analysis-synthesis framework based on these modern generative models, enhanced by recent advances in machine learning. We first review existing approaches, both in sound synthesis and in generative machine learning, and focus on how our work inserts itself in both practices and what can be expected from their collation. Subsequently, we focus a little more on generative models, and how modern advances in the domain can be exploited to allow us learning complex sound distributions, while being sufficiently flexible to be integrated in the creative flow of the user. We then propose an inference / generation process, mirroring analysis/synthesis paradigms that are natural in the audio processing domain, using latent models that are based on a continuous higher-level space, that we use to control the generation. We first provide preliminary results of our method applied on spectral information, extracted from several datasets, and evaluate both qualitatively and quantitatively the obtained results. Subsequently, we study how to make these methods more suitable for learning audio data, tackling successively three different aspects. First, we propose two different latent regularization strategies specifically designed for audio, based on and signal / symbol translation and perceptual constraints. Then, we propose different methods to address the inner temporality of musical signals, based on the extraction of multi-scale representations and on prediction, that allow the obtained generative spaces that also model the dynamics of the signal. As a last chapter, we swap our scientific approach to a more research & creation-oriented point of view: first, we describe the architecture and the design of our open-source library, vsacids, aiming to be used by expert and non-expert music makers as an integrated creation tool. Then, we propose an first musical use of our system by the creation of a real-time performance, called aego, based jointly on our framework vsacids and an explorative agent using reinforcement learning to be trained during the performance. Finally, we draw some conclusions on the different manners to improve and reinforce the proposed generation method, as well as possible further creative applications.
Tra i diversi campi di ricerca nell’ambito dell’informatica musicale, la sintesi e la generazione di segnali audio incarna la pluridisciplinalità di questo settore, nutrendo insieme le pratiche scientifiche e musicale dalla sua creazione. Inerente all’informatica dalla sua creazione, la generazione audio ha ispirato numerosi approcci, evolvendo colle pratiche musicale e gli progressi tecnologici e scientifici. Inoltre, alcuni processi di sintesi permettono anche il processo inverso, denominato analisi, in modo che i parametri di sintesi possono anche essere parzialmente o totalmente estratti dai suoni, dando una rappresentazione alternativa ai segnali analizzati. Per di più, la recente ascesa dei algoritmi di l’apprendimento automatico ha vivamente interrogato il settore della ricerca scientifica, fornendo potenti data-centered metodi che sollevavano diversi epistemologici interrogativi, nonostante i sui efficacia. Particolarmente, un tipo di metodi di apprendimento automatico, denominati modelli generativi, si concentrano sulla generazione di contenuto originale usando le caratteristiche che hanno estratti dei dati analizzati. In tal caso, questi modelli non hanno soltanto interrogato i precedenti metodi di generazione, ma anche sul modo di integrare questi algoritmi nelle pratiche artistiche. Mentre questi metodi sono progressivamente introdotti nel settore del trattamento delle immagini, la loro applicazione per la sintesi di segnali audio e ancora molto marginale. In questo lavoro, il nostro obiettivo e di proporre un nuovo metodo di audio sintesi basato su questi nuovi tipi di generativi modelli, rafforazti dalle nuove avanzati dell’apprendimento automatico. Al primo posto, facciamo una revisione dei approcci esistenti nei settori dei sistemi generativi e di sintesi sonore, focalizzando sul posto di nostro lavoro rispetto a questi disciplini e che cosa possiamo aspettare di questa collazione. In seguito, studiamo in maniera più precisa i modelli generativi, e come possiamo utilizzare questi recenti avanzati per l’apprendimento di complesse distribuzione di suoni, in un modo che sia flessibile e nel flusso creativo del utente. Quindi proponiamo un processo di inferenza / generazione, il quale rifletta i processi di analisi/sintesi che sono molto usati nel settore del trattamento del segnale audio, usando modelli latenti, che sono basati sull’utilizzazione di un spazio continuato di alto livello, che usiamo per controllare la generazione. Studiamo dapprima i risultati preliminari ottenuti con informazione spettrale estratte da diversi tipi di dati, che valutiamo qualitativamente e quantitativamente. Successiva- mente, studiamo come fare per rendere questi metodi più adattati ai segnali audio, fronteggiando tre diversi aspetti. Primo, proponiamo due diversi metodi di regolarizzazione di questo generativo spazio che sono specificamente sviluppati per l’audio : una strategia basata sulla traduzione segnali / simboli, e una basata su vincoli percettivi. Poi, proponiamo diversi metodi per fronteggiare il aspetto temporale dei segnali audio, basati sull’estrazione di rappresentazioni multiscala e sulla predizione, che permettono ai generativi spazi ottenuti di anche modellare l’aspetto dinamico di questi segnali. Per finire, cambiamo il nostro approccio scientifico per un punto di visto piú ispirato dall’idea di ricerca e creazione. Primo, descriviamo l’architettura e il design della nostra libreria open-source, vsacids, sviluppata per permettere a esperti o non-esperti musicisti di provare questi nuovi metodi di sintesi. Poi, proponiamo una prima utilizzazione del nostro modello con la creazione di una performance in real- time, chiamata ægo, basata insieme sulla nostra libreria vsacids e sull’uso di une agente di esplorazione, imparando con rinforzo nel corso della composizione. Finalmente, tramo dal lavoro presentato alcuni conclusioni sui diversi modi di migliorare e rinforzare il metodo di sintesi proposto, nonché eventuale applicazione artistiche.
MANIFOLD REPRESENTATIONS OF MUSICAL SIGNALS AND GENERATIVE SPACES
CHEMLA ROMEU SANTOS, AXEL CLAUDE ANDRE'
2020
Abstract
Among the diverse research fields within computer music, synthesis and generation of audio signals epitomize the cross-disciplinarity of this domain, jointly nourishing both scientific and artistic practices since its creation. Inherent in computer music since its genesis, audio generation has inspired numerous approaches, evolving both with musical practices and scientific/technical advances. Moreover, some syn- thesis processes also naturally handle the reverse process, named analysis, such that synthesis parameters can also be partially or totally extracted from actual sounds, and providing an alternative representation of the analyzed audio signals. On top of that, the recent rise of machine learning algorithms earnestly questioned the field of scientific research, bringing powerful data-centred methods that raised several epistemological questions amongst researchers, in spite of their efficiency. Especially, a family of machine learning methods, called generative models, are focused on the generation of original content using features extracted from an existing dataset. In that case, such methods not only questioned previous approaches in generation, but also the way of integrating this methods into existing creative processes. While these new generative frameworks are progressively introduced in the domain of image generation, the application of such generative techniques in audio synthesis is still marginal. In this work, we aim to propose a new audio analysis-synthesis framework based on these modern generative models, enhanced by recent advances in machine learning. We first review existing approaches, both in sound synthesis and in generative machine learning, and focus on how our work inserts itself in both practices and what can be expected from their collation. Subsequently, we focus a little more on generative models, and how modern advances in the domain can be exploited to allow us learning complex sound distributions, while being sufficiently flexible to be integrated in the creative flow of the user. We then propose an inference / generation process, mirroring analysis/synthesis paradigms that are natural in the audio processing domain, using latent models that are based on a continuous higher-level space, that we use to control the generation. We first provide preliminary results of our method applied on spectral information, extracted from several datasets, and evaluate both qualitatively and quantitatively the obtained results. Subsequently, we study how to make these methods more suitable for learning audio data, tackling successively three different aspects. First, we propose two different latent regularization strategies specifically designed for audio, based on and signal / symbol translation and perceptual constraints. Then, we propose different methods to address the inner temporality of musical signals, based on the extraction of multi-scale representations and on prediction, that allow the obtained generative spaces that also model the dynamics of the signal. As a last chapter, we swap our scientific approach to a more research & creation-oriented point of view: first, we describe the architecture and the design of our open-source library, vsacids, aiming to be used by expert and non-expert music makers as an integrated creation tool. Then, we propose an first musical use of our system by the creation of a real-time performance, called aego, based jointly on our framework vsacids and an explorative agent using reinforcement learning to be trained during the performance. Finally, we draw some conclusions on the different manners to improve and reinforce the proposed generation method, as well as possible further creative applications.File | Dimensione | Formato | |
---|---|---|---|
phd_unimi_R11707.pdf
accesso aperto
Dimensione
36.07 MB
Formato
Adobe PDF
|
36.07 MB | Adobe PDF | Visualizza/Apri |
I documenti in UNITESI sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/20.500.14242/82430
URN:NBN:IT:UNIMI-82430