Spherical harmonics domain-based approach for source localization in presence of directional interference

Area of Science:

• Acoustics
• Signal Processing
• Machine Learning

Background:

• Accurate source localization is crucial in various applications, but challenging under reverberant and noisy conditions with directional interference.
• Existing methods often struggle to effectively mitigate interference while preserving target source information.

Purpose of the Study:

• To develop a robust learning-based method for source localization in challenging acoustic environments.
• To effectively filter directional interference using an attention mechanism prior to localization.

Main Methods:

• Spherical harmonic decomposition of microphone array recordings to extract spherical harmonics coefficients.
• Integration of an attention mechanism via a binary mask to filter undesired source components.
• Training a convolutional neural network to map filtered coefficients' phase and magnitude to location classes.

Main Results:

• Demonstrated effective filtering of dominant undesired source components.
• Achieved accurate source localization by mapping filtered spherical harmonics coefficients.
• The proposed method shows promise for real-world applications with interference.

Conclusions:

• The developed learning-based method successfully addresses source localization challenges posed by directional interference.
• The attention-based filtering mechanism is key to improving localization accuracy in adverse conditions.
• This approach offers a significant advancement in robust acoustic signal processing.