Slow and steady: auditory features for discriminating animal vocalizations

Area of Science:

• Bioacoustics
• Auditory Perception
• Machine Learning

Background:

• Animal vocalizations contain complex spectro-temporal information.
• Listeners may leverage temporal regularities for sound discrimination.

Purpose of the Study:

• To investigate if temporal regularities in vocalizations aid in species and individual discrimination.
• To identify the key acoustic features responsible for vocalization recognition.

Main Methods:

• Utilized Slow Feature Analysis (SFA) to extract temporally regular features from vocalizations.
• Included vocalizations from birds, humans, and rhesus macaques.
• Employed a trained classifier to test auditory discrimination based on learned features.

Main Results:

• Achieved excellent vocalization discrimination (>95%) across all tested groups.
• Discrimination performance was primarily driven by approximately 10 highly temporally regular features.
• Identified that these regular features correlate strongly with predictable sound components.

Conclusions:

• Temporal regularities are crucial for auditory discrimination of animal vocalizations.
• A small set of temporally regular features effectively captures essential information for sound recognition.
• The findings support the hypothesis that predictable acoustic patterns are fundamental for vocalization processing.