Perceptually motivated wavelet packet transform for bioacoustic signal enhancement

Area of Science:

• Bioacoustics
• Signal Processing
• Machine Learning

Background:

• Background noise in bioacoustic recordings is a significant challenge.
• Adverse recording environments degrade signal quality.
• Effective noise reduction is crucial for species identification and behavior analysis.

Purpose of the Study:

• To propose a novel bioacoustic signal enhancement technique.
• To improve the quality of bioacoustic signals degraded by background noise.
• To offer a versatile method applicable to a wide range of species.

Main Methods:

• A perceptually scaled wavelet packet decomposition is employed.
• A species-specific Greenwood scale function is utilized.
• Spectral estimation techniques adapted from human speech enhancement are applied.

Main Results:

• The proposed method significantly outperforms traditional techniques like bandpass filtering, spectral subtraction, Wiener filtering, and Ephraim-Malah filtering.
• Evaluations across multiple species (bird, primate, whale) and noise types demonstrate superior performance.
• Enhanced signal-to-noise ratio (SNR) and segmental SNR confirm the effectiveness of the approach.

Conclusions:

• The novel bioacoustic enhancement technique offers superior performance in noisy environments.
• This method provides a robust solution for improving the quality of bioacoustic recordings.
• The approach is effective across diverse species and noise conditions, advancing bioacoustic research.