ARTIFICIAL intelligence may soon play a role in how whip-use rules are enforced in horseracing, according to new research. The study, led by Japanese researchers, investigates whether whip strikes can be reliably detected using high-resolution audio recordings analysed by advanced deep-learning systems.

Currently, enforcement of whip regulations relies heavily on stewards manually reviewing race footage, a process that is time-consuming and can be open to interpretation. The researchers set out to explore whether sound, rather than video alone, could provide a more objective and automated method of identifying whip strikes during races.

Acoustic details

A key finding of the study is that whip strikes produce extremely short, sharp sounds containing very high-frequency components, many of which are lost in standard audio recordings. To capture these acoustic details, the researchers used microphones sampling at 192 kHz, more than four times the rate of typical consumer audio. This allowed the system to retain the subtle spectral features that distinguish whip strikes from background noise, such as hoof beats, crowd reactions and wind.

Recurrent layers

The team then applied a form of artificial intelligence well-suited to analysing time-based data like sound. The “convolutional layers” extract distinctive audio features, while recurrent layers analyse how those features evolve over time, a crucial capability when dealing with fleeting events that last only milliseconds.

One of the challenges the researchers addressed was class imbalance: whip strikes are relatively rare events within long stretches of race audio. The study tested several strategies to ensure the AI did not simply learn to ignore these uncommon but important sounds.

When evaluated using audio from 24 official races in Japan, the best-performing model correctly detected around 70% of 620 confirmed whip strikes.

While this accuracy is not yet sufficient to replace human judgement, the authors stress that it represents a significant step forward given the complexity of real-world racing environments.

Live monitoring

Notably, the system was capable of operating close to real time, raising the possibility of live monitoring in the future.

From an equine welfare perspective, the research is particularly relevant. Objective, data-driven tools could help racing authorities apply whip regulations more consistently and transparently, while also providing clearer evidence in disciplinary cases.

The researchers conclude that, with further refinement, particularly in noisy race conditions, audio-based AI systems could become a valuable addition to stewarding practices, reinforcing both fair competition and the welfare of horses at the heart of the sport.