Roar Data: Redefining a Lion's Roar Using Machine Learning.

Area of Science:

• Bioacoustics
• Wildlife Conservation
• Animal Communication

Background:

• African lions use roaring bouts for territorial and social communication.
• Individual lion roars are unique identifiers, making them valuable for population studies.
• Current methods for identifying roars rely on expert inference, introducing bias.

Purpose of the Study:

• To develop a data-driven approach for automatically classifying African lion full-throated roars.
• To differentiate between various vocalizations within a lion's roaring bout.
• To enhance the accuracy of individual identification and population density estimation using acoustic data.

Main Methods:

• Utilized two-state Gaussian Hidden-Markov Models for roar classification.
• Employed K-means clustering with acoustic metrics (frequency, duration) for call type classification.
• Compared data-driven roar classification with manual methods for individual identification.

Main Results:

• Successfully classified two types of lion roars (full-throated and intermediary) with 84.7% accuracy.
• Achieved 95.4% accuracy in classifying lion call types using simple acoustic metrics and K-means clustering.
• Data-driven classification improved individual identification (F1-score 0.87) compared to manual classification (0.80).

Conclusions:

• A novel, accessible data-driven method for classifying lion roars has been established.
• This approach reduces human bias in acoustic monitoring and enhances individual identification.
• The findings facilitate the use of passive acoustic monitoring for lion population studies, complementing traditional methods.