In this thesis, the profound intersection of deep learning and audio processing is explored, highlighting the transformative potential of these techniques in deciphering and manipulating audio signals. From the intricacies of marine ecosystems to the nuances of music, the application of deep learning has shown considerable promise in reshaping our understanding of sound. We commence by delving into deep extractors for audio source separation, showcasing their potency in tasks ranging from isolating marine sounds to identifying singing voices in music tracks. This journey emphasizes the role of neural networks in extracting and interpreting sounds from a complex mixture of signals, taking us to the world of autoregressive models, where we investigate their principles and applications in source separation, emphasizing unsupervised methods. Much of the research dwells on the innovative Bayesian approach with autoregressive models for signal source separation, demonstrating its efficiency across auditory and visual domains and the intriguing application of diffusion models for music generation and separation, accentuating their versatility in audio tasks. While the technical profundities form the core of the research, its broader implications shed light on the transformative potential of deep learning in myriad domains. From music production and music information retrieval to environmental surveillance, the adaptability of deep learning techniques promises a future replete with sophisticated audio processing tools. Conclusively, this thesis stands as a testament to the power of deep learning in enhancing, understanding, and enriching the world of sound, paving the way for further advancements in this captivating realm.

Harmonizing deep learning: a journey through the innovations in signal processing, source separation and music generation / Mancusi, Michele. - (2024 Jan 26).

Harmonizing deep learning: a journey through the innovations in signal processing, source separation and music generation

MANCUSI, MICHELE
26/01/2024

Abstract

In this thesis, the profound intersection of deep learning and audio processing is explored, highlighting the transformative potential of these techniques in deciphering and manipulating audio signals. From the intricacies of marine ecosystems to the nuances of music, the application of deep learning has shown considerable promise in reshaping our understanding of sound. We commence by delving into deep extractors for audio source separation, showcasing their potency in tasks ranging from isolating marine sounds to identifying singing voices in music tracks. This journey emphasizes the role of neural networks in extracting and interpreting sounds from a complex mixture of signals, taking us to the world of autoregressive models, where we investigate their principles and applications in source separation, emphasizing unsupervised methods. Much of the research dwells on the innovative Bayesian approach with autoregressive models for signal source separation, demonstrating its efficiency across auditory and visual domains and the intriguing application of diffusion models for music generation and separation, accentuating their versatility in audio tasks. While the technical profundities form the core of the research, its broader implications shed light on the transformative potential of deep learning in myriad domains. From music production and music information retrieval to environmental surveillance, the adaptability of deep learning techniques promises a future replete with sophisticated audio processing tools. Conclusively, this thesis stands as a testament to the power of deep learning in enhancing, understanding, and enriching the world of sound, paving the way for further advancements in this captivating realm.
26-gen-2024
File allegati a questo prodotto
File Dimensione Formato  
Tesi_dottorato_Mancusi.pdf

accesso aperto

Note: Tesi completa
Tipologia: Tesi di dottorato
Licenza: Creative commons
Dimensione 18.67 MB
Formato Adobe PDF
18.67 MB Adobe PDF

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11573/1701262
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus ND
  • ???jsp.display-item.citation.isi??? ND
social impact