PROTAX-Sound: A probabilistic framework for automated animal sound identification

Area of Science:

• Bioacoustics
• Computational Ecology
• Machine Learning

Background:

• Autonomous audio recording is revolutionizing bioacoustics for cost-effective species surveys.
• A key challenge is developing reliable classifiers for multi-species identification from sound data.

Purpose of the Study:

• To present PROTAX-Sound, a statistical framework for probabilistic classification of animal sounds.
• To improve multi-species identification accuracy in bioacoustic monitoring.

Main Methods:

• PROTAX-Sound utilizes a multinomial regression model, integrating various sound features and existing algorithm outputs.
• It employs audio and image processing to identify regions of interest, extract acoustic features, and compare them against a reference database.
• The framework provides probabilistic classifications, including the potential for identifying species outside the reference database.

Main Results:

• The PROTAX-Sound framework demonstrated effective classification of animal sounds.
• In a tropical bird case study, the best performing classifier achieved 68% accuracy for 200 species.
• The system combines information from multiple classifiers to produce calibrated classification probabilities, enhancing accuracy.

Conclusions:

• PROTAX-Sound significantly improves upon current techniques for classifying animal vocalizations.
• This framework offers a promising solution for accurate and efficient multi-species identification in bioacoustics.
• Its probabilistic output and ability to identify novel species enhance its utility for ecological research.