Image by Marta Fernandez Jimenez/Shutterstock

Regulations limit both the intensity and frequency of whip use during horse racing. Nevertheless, compliance is currently verified manually after each race. Researchers at University of Tsukuba have developed an innovative system by combining high-resolution audio recording with artificial intelligence for automatic detection of whip sounds. This technology improves the analysis of high-frequency components, significantly enhancing the feasibility of real-time judgment.

Tsukuba, Japan—Horse jockeys traditionally use whips to encourage acceleration and maintain the focus of the horse. However, applying excessive force or exceeding the permitted number of strikes violates racing rules and raises concerns about animal welfare and fairness. Currently, race stewards check whip use by manually reviewing video footage of the race. This time-consuming, error-prone process would greatly benefit from automated detection technologies.

This research focuses on the distinctive sound of whip swings. Because whip sounds are extremely brief and contain very-high-frequency components, they cannot be accurately captured in standard audio recordings. Accordingly, the research team employed high-resolution audio recording at 192 kHz and trained a deep learning model—a convolutional recurrent neural network—to learn the acoustic features and temporal variations of whip strikes.

The best-performing model accurately detected approximately 70% of 620 annotated whip strikes in audio data collected from 24 official races in Japan. Importantly, the study demonstrated that high-frequency components are critical for accurate detection, providing the first confirmation of "very high-pitched" elements in whip sounds. Furthermore, the system achieved faster-than-real-time audio processing under many conditions, indicating its strong potential for live race monitoring.

This research is expected to promote fair competition and improve animal welfare by ensuring the appropriate use of whips. In future work, the dataset will be extended and the system robustness enhanced under noisy conditions. The ultimate goal is practical implementation in real-world racing environments.

###
This research was supported by NEXION Corporation.

Original Paper

Title of original paper:

Whip strike detection using high-sampling-rate audio by evaluating convolutional recurrent neural network configurations and class imbalance

Journal:

Engineering Applications of Artificial Intelligence

DOI:

10.1016/j.engappai.2025.113272

Correspondence

Associate Professor ZEMPO Keiichi
Institute of Systems and Information Engineering, University of Tsukuba

Related Link

Institute of Systems and Information Engineering