The human ability to move in time with music, known as auditory–motor synchronisation, is a complex process whose development is not fully understood. Scientists at McMaster University have proposed a model based on a recurrent neural network trained using reinforcement learning methods, which exhibits dynamics similar to those of humans when synchronising with a beat. The study, published on 12 September 2026 in the journal Communications Psychology, sheds light on the mechanisms underlying this phenomenon.
Modelling synchronisation with a beat
The study involved creating a recurrent neural network that receives a sequence of metronome signals at various tempos as input and generates a series of ‘finger snaps’ with a slight delay. The network was trained under four different schemes, each of which encouraged synchrony between the snaps and signals in its own way. The most successful scheme rewarded earlier snaps more than later ones and also encouraged matching the tempo.
The trained agent demonstrated an ability to correct synchronisation errors that matches findings from studies of human behaviour both qualitatively and quantitatively. An asymmetry in correction was observed: lagging behind is corrected more strongly than moving ahead. When timing fluctuations (jitter) were introduced during training or testing, the agent tended to anticipate the signals (‘negative mean asynchrony’), a tendency that closely resembles human findings and scales with the level of fluctuation and the duration of training.
An internal sense of rhythm and imitation
The model also showed an ability to maintain precisely timed sequences of snaps during the continuation phase, demonstrating that it had internalised the tempo, even though this aspect was not included in training. Neural activity in the model revealed patterns observed in primates trained to synchronise, as well as signs of subjective rhythmisation—the human tendency to perceive metronome clicks in groups of two.
The study’s results suggest that intrinsic reinforcement for early actions and imitation may play an important role in the development of auditory–motor synchronisation. The model provides a system for further study of the mechanisms involved in learning to synchronise.
