Deep learning assisted sound source localization using two orthogonal first-order differential microphone arrays.

Area of Science:

• Acoustics
• Signal Processing
• Machine Learning

Background:

• Sound source localization in noisy and reverberant environments is difficult, particularly with small microphone arrays.
• Deep learning methods offer promising solutions by treating localization as a classification task.
• Existing methods often rely on Generalized Cross-Correlation Phase Transform (GCC-PHAT) features, which may not perform well with small arrays.

Purpose of the Study:

• To propose a deep learning-assisted sound localization method for small-sized microphone arrays.
• To develop an improved feature extraction scheme for enhanced robustness in adverse acoustic conditions.
• To evaluate the proposed method against state-of-the-art techniques.

Main Methods:

• Utilized a small-sized microphone array composed of two orthogonal first-order differential microphone arrays.
• Developed an improved feature extraction method based on sound intensity estimation.
• Decoupled sound pressure and particle velocity components in whitening weighting for enhanced feature robustness.
• Employed deep learning for sound source localization.

Main Results:

• The proposed deep learning approach achieved higher spatial resolution compared to existing methods.
• The method demonstrated superior performance in noisy and reverberant environments for small arrays.
• Simulation and real-world experiments validated the effectiveness of the proposed technique.

Conclusions:

• The proposed deep learning-assisted sound localization method is effective for small microphone arrays.
• The improved sound intensity-based feature extraction enhances robustness in challenging acoustic conditions.
• This approach offers a significant advancement over GCC-PHAT and conventional sound intensity features for small-array applications.