Source localization with acoustic sensor arrays using generative model based fitting with sparse constraints.

Area of Science:

• Acoustics
• Signal Processing
• Machine Learning

Background:

• Acoustic source localization is crucial for applications like robotics and surveillance.
• Existing Steered Response Power (SRP) methods can be sensitive to noise and model inaccuracies.
• Accurate localization in reverberant indoor environments remains a challenge.

Purpose of the Study:

• To develop a novel and robust approach for indoor acoustic source localization.
• To improve the accuracy of acoustic source position estimation using sensor arrays.
• To address limitations of traditional SRP methods in real-world acoustic conditions.

Main Methods:

• A generative model is defined to represent acoustic power maps from SRP data.
• An optimization framework is proposed to fit the generative model to real SRP data.
• Sparse constraints and subspace analysis are employed to enhance model fitting and signal filtering.

Main Results:

• The proposed method demonstrates statistically significant reductions in localization error.
• Error reductions of up to 30% were achieved compared to SRP-PHAT strategies.
• Experimental validation was performed using a realistic speech database.

Conclusions:

• The novel generative model-based approach offers improved performance for indoor acoustic source localization.
• The integration of sparse constraints and subspace analysis enhances robustness against noise and unmodelled effects.
• This work provides a more accurate and reliable solution for acoustic source localization in complex environments.