Neural network for multi-exponential sound energy decay analysis.

Area of Science:

• Acoustics
• Signal Processing
• Machine Learning

Background:

• Established sound energy decay function (EDF) models involve superpositions of exponentials and noise.
• Accurate parameter estimation is crucial for understanding acoustic environments.

Purpose of the Study:

• To propose a novel neural-network-based approach for estimating parameters of sound energy decay functions (EDFs).
• To evaluate the network's performance on extensive real-world acoustic measurement datasets.

Main Methods:

• Development of a lightweight neural network architecture.
• Training the network on synthetic EDFs.
• Evaluation using two large datasets comprising over 20,000 measured EDFs from diverse acoustic environments.

Main Results:

• The neural network robustly estimates EDF model parameters.
• The proposed method demonstrates computational efficiency.
• The network performs well across various acoustic conditions.

Conclusions:

• The neural network offers a reliable and efficient solution for EDF parameter estimation.
• The approach is suitable for large-scale acoustic data analysis.
• Public availability of the implementation facilitates broader adoption.