Full text
| |
| Author(s): |
Fernanda Caldas Correia
Total Authors: 1
|
| Document type: | Master's Dissertation |
| Press: | Campinas, SP. |
| Institution: | Universidade Estadual de Campinas (UNICAMP). Faculdade de Engenharia Elétrica e de Computação |
| Defense date: | 2021-07-07 |
| Examining board members: |
Bruno Sanches Masiero;
Romis Ribeiro de Faissol Attux;
Eric Brandão Carneiro
|
| Advisor: | Bruno Sanches Masiero |
| Abstract | |
When doing spatial audio processing of sound scenes it is often necessary to first detect the sound sources present in the scene, which is commonly done with the aid of a microphone array and a direction of arrival (DOA) estimation algorithm. If the system should be able to analyze sound coming from all possible directions, then spherical microphone arrays are most commonly used. Classical DOA algorithms, such as plane-wave decomposition or spherical beamforming, suffer from low localization accuracy. In an attempt to improve DOA estimation, the compressive beamforming (CB) algorithm has been proposed. CB applies sparsity regularization to the localization problem through the use of L1-norm minimization, therefore taking into account the assumption that common sound scenes are usually composed of only a handful of sound sources. In this thesis, the performance of three sparsity regularization algorithms on a plane-wave decomposition model is compared: the $\ell_1$-norm minimization via Disciplined Convex Program (DCP), the Least Absolute Shrinkage and Selection Operator (LASSO) method and the Orthogonal Matching Pursuit (OMP). It is shown that the three algorithms were able to accurately estimate the number of sources and its directions for an artificial sound scene, both with and without noise. However, the performance was shown to deteriorated when applied at a practical situation with a sound source recorded in an anechoic chamber. In this case, there was an improvement with the combination of the LASSO to estimate the number of sound sources and the OMP to refine the wave amplitude (AU) | |
| FAPESP's process: | 18/26531-6 - Study of sparsity-promoting regularization for superdirectional microphone with spherical geometry |
| Grantee: | Fernanda Caldas Correia |
| Support Opportunities: | Scholarships in Brazil - Master |