Machine-learning-based simultaneous detection and ranging of impulsive baleen whale vocalizations using a single...

Area of Science:

• Marine bioacoustics
• Signal processing
• Machine learning

Background:

• Baleen whale gunshot vocalizations propagate dispersively in shallow waters, modeled by normal mode theory.
• Traditional underwater acoustic source ranging requires complex, costly multi-hydrophone arrays.
• Existing single-hydrophone methods for whale vocalization ranging need manual signal detection and labor.

Purpose of the Study:

• To develop a temporal convolutional network (TCN) for simultaneous detection and ranging of baleen whale gunshot vocalizations using single-hydrophone data.
• To automate and improve the efficiency of whale vocalization analysis in shallow marine environments.

Main Methods:

• Applied a TCN to spectrograms of single-hydrophone acoustic data.
• Trained the TCN using simulated gunshot data across various environments and ranges, incorporating experimental noise.
• Validated the TCN on North Pacific right whale gunshot data from the Bering Sea, comparing results to a time-warping inversion method.

Main Results:

• The TCN successfully detected gunshot vocalizations among noise with high precision.
• TCN-estimated ranges closely matched those obtained from the established time-warping inversion method.
• The TCN achieved joint detection and ranging, reducing the need for preliminary signal processing and human intervention.

Conclusions:

• The TCN offers an efficient and accurate method for detecting and ranging baleen whale gunshot vocalizations from single-hydrophone data.
• This approach simplifies acoustic monitoring and has potential for estimating shallow-water geoacoustic properties.
• The TCN demonstrates a promising advancement in automated marine bioacoustic analysis.