A Bayesian direction-of-arrival model for an undetermined number of sources using a two-microphone array

Area of Science:

• Acoustics and Signal Processing
• Computational Statistics

Background:

• Sound source localization is crucial for applications like robotics and video conferencing.
• Current methods often require prior knowledge of the number of sound sources.
• Estimating source location typically relies on time-difference-of-arrival (TDOA) signals.

Purpose of the Study:

• To present an updated method for sound source localization that infers both the number and position of sources.
• To demonstrate the effectiveness of nested sampling with a Laplacian mixture model for speech localization.
• To compare the proposed method against established sampling techniques.

Main Methods:

• Utilized nested sampling to explore a probability distribution of source positions.
• Employed a Laplacian mixture model to infer source characteristics.
• Evaluated the method through various experimental setups and scenarios.

Main Results:

• The nested sampling approach accurately determined the number and location of speech sources.
• Experimental results validated the proposed method's viability.
• Demonstrated superior performance compared to popular sampling methods in speech localization tasks.

Conclusions:

• Nested sampling is an accurate and effective tool for sound source localization.
• The Laplacian mixture model enhances the ability to infer multiple source parameters.
• This method offers a robust solution for scenarios where source count is unknown.