.png){kind=icon}
.png){kind=icon}
Return
Microphone array based joint optimization framework combining GSC and LDA for outdoor multi-source recognition
.png){kind=icon}
PRE
AI
DOI：10.1016/j.sigpro.2026.110519.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Sound source enhancement (SSE) techniques have been increasingly adopted as pre-processors for sound source recognition (SSR). While joint optimization methods outperform conventional simply-cascading approaches by resolving the mismatch in objectives between SSE and SSR modules, existing solutions still face some key limitations. They rely predominantly on data-intensive neural networks, restricting their applicability in outdoor multi-source scenarios characterized by limited training data and random directions of arrival (DOAs). Furthermore, the potential of microphone array has not yet been sufficiently exploited from an array signal processing perspective. To address these issues, this article proposes a novel microphone array based joint optimization framework for outdoor multi-source recognition. As for front-end SSE module, the traditional generalized sidelobe canceller (GSC) is modified by introducing a joint optimization matrix (JOM) to receive feedback from SSR module. Meanwhile, the linear discriminant analysis (LDA) is employed in the SSR module to maximize between-class separability. By coupling the JOM with the projection matrix of LDA, the SSE is guided to produce enhanced outputs that preserve recognition-critical features, rather than merely maximizing the signal-to-interference-plus-noise ratio (SINR) output. Extensive evaluations demonstrate that the proposed approach achieves superior recognition performance compared to state-of-the-art alternatives, thus verifying its effectiveness in outdoor multi-source scenarios.
Journal
IF:
3.6
Papers:
9.9K
Citations:
1.7W