30/09/2026
Finding whale calls in days of underwater audio takes time, especially when background noise masks the sounds researchers want to study.
For his Electrical and Electronic Engineering project at Stellenbosch University, Jonathan Rule, under the supervision of Professor Jaco Versfeld, developed software to detect and classify Bryde’s whale vocalisations in recordings from False Bay. It combines audio processing with hidden Markov models, which use patterns over time to identify likely calls.
In a two-minute evaluation clip, the detection configuration found 28 of 30 labelled calls and produced six false alarms. A separate classification configuration assigned the correct call type to 26 calls, misclassified four, and produced seven false alarms. A detection counted as correct when it overlapped any part of a labelled call, so these figures don’t establish accurate call boundaries.
The software exports suggested labels to Raven Pro for researchers to review. Further testing on longer recordings and rare call types is needed to assess its use in broader monitoring.
Read the full article for the approach, findings, and limitations: https://www.su.ac.za/en/faculties/engineering/departments/electrical-electronic-engineering/news/automated-detection-and-classification-brydes-whale-vocalisations
