Acoustic Impulsive Noise Based on Non-Gaussian Models: An Experimental Evaluation.

Area of Science:

• Signal Processing
• Acoustics
• Statistical Modeling

Background:

• Acoustic channels often exhibit non-Gaussian and non-stationary characteristics.
• Traditional signal processing methods assuming Gaussian distributions perform poorly on such data.
• Impulsive noise significantly degrades audio signal quality and analysis.

Purpose of the Study:

• To analyze audio signals corrupted by impulsive noise using non-Gaussian models.
• To evaluate the suitability of different statistical models for audio signal processing.
• To identify the optimal model for representing audio signals with impulsive noise.

Main Methods:

• Comparison of audio samples against Gaussian, alpha-stable, and Gaussian mixture models.
• Evaluation of model fitting using graphical and numerical methods.
• Analysis of fitting properties, including window length and overlap.

Main Results:

• The Gaussian model provided a poor fit for audio signals with impulsive noise.
• Gaussian mixture models showed moderate fitting capabilities.
• The alpha-stable model demonstrated the best fit across all tested audio scenarios.

Conclusions:

• Non-Gaussian models are essential for accurate audio signal processing in the presence of impulsive noise.
• The alpha-stable distribution is a superior model for characterizing impulsive noise in acoustic channels.
• Findings suggest a need to revise signal processing techniques to incorporate non-Gaussian assumptions for improved performance.