Diego de Benito
AI Engineer & Researcher
Madrid, Spain PhD in Computer Science and Telecommunications AI Specialist at EVO Banco
Ph.D. Thesis: Multi-resolution and Source Separation for Improved Sound Event Detection based on Deep Neural Networks
- 🔗 Link to Ph.D. Dissertation
- Defense Date: October 9th, 2023
- Research Group: Audias - Audio, Data Intelligence and Speech
- Institution: Universidad Autónoma de Madrid
- Grade: Cum Laude
The slides for the dissertation are available here.
During my Ph.D., I focused on advancing Sound Event Detection (SED) using deep learning techniques. My research involved developing novel methods to enhance audio signal representations, particularly through a multi-resolution approach that captures diverse time and frequency characteristics of sound events. I participated in the DCASE Challenge, improving SED performance each year. Additionally, I explored the use of Source Separation neural networks to preprocess audio mixtures, leading to cleaner signals and better SED outcomes.
My work also included proposing a new distractor measure, studying mel-spectrogram resolutions, creating a synthetic dataset for severe event overlap, and analyzing semi-supervised training methods. These contributions aimed to address challenging SED scenarios, such as acoustic degradation and event overlap, across various sound event categories.