🔬 Research Topics & Projects
My research revolves around Audio Signal Processing, Machine Learning, and Music Applications, with a strong emphasis on source separation and sound analysis.
1. Advanced Source Separation and Audio Analysis
A significant portion of my work centers on developing and improving Non-Negative Matrix Factorization (NMF) and related algorithms for separating sound components.
Non-Negative Matrix Factorization (NMF): Extensive application and innovation around various NMF models, including Multichannel NMF, Constrained NMF, Orthogonal NMF, and Directivity-Aware NMF.
Computational Efficiency: Developing methods, such as an efficient parallel kernel based on Cholesky decomposition, to accelerate multichannel NMF algorithms for high-performance computing.
Spatial Priors: Incorporating pre-trained spatial priors on Multichannel NMF for better source separation performance.
2. Music Information Retrieval (MIR) and Audio Applications
My work in music technology focuses on analyzing and manipulating musical signals, stemming from your PhD dissertation on polyphonic music transcription.
Music Source Separation: Developing systems for separating musical elements (like vocals and instruments), including Ambisonics domain Singing Voice Separation, often combining Deep Neural Networks (DNNs) with Direction Aware Multichannel NMF.
Polyphonic Music Transcription: Your PhD work was focused on Musical Instruments Model Estimation for Polyphonic Music Transcription.
Score-Informed Processing: Researching methods for audio-to-score alignment at the note level and improving source separation for classical music by refining note information.
3. Biomedical Signal Processing (Health Applications)
My research here is focused on signal processing and machine learning techniques to the analysis and detection of pathological body sounds, often utilizing NMF for noise reduction and feature extraction.
Respiratory Sound Analysis: Developing robust systems for the automatic detection, classification, and localization of adventitious sounds like wheezing and crackle using:
Constrained Low-Rank NMF and semi-supervised NMF approaches.
Convolutional Neural Networks (CNNs) based on cochleograms.
Incremental NMF algorithms for ambient denoising in auscultation.
Heart Sound Analysis: Detection of valvular heart diseases by combining Orthogonal NMF and Convolutional Neural Networks (CNNs) in PCG signals.
Snore Detection: Improving snore detection under limited datasets using harmonic/percussive source separation and CNNs.
4. Spatial Audio and Acoustics
Techniques for analyzing and separating sounds in three-dimensional space.
Sound Field Decomposition: Researching Spherical-harmonics-based sound field decomposition for sound source separation.
Ray-Space NMF: Applying Ray-Space-Based Multichannel NMF for advanced audio source separation.
Educational Tools: Development of a 3D Audio Interactive Simulator for experiential learning of spatial audio techniques.
Specific Funded Project
Researching and Encouraging the Promulgation of European Repertory through Technologies Operating on Records Interrelated Utilising Machines (Project grant from the European Commission).