PREDICTIVE AI FOR RHYTHM SYNCHRONIZATION IN TRAINING

Abstract [English]

Rhythm synchronization is a predictive type of AI that builds upon temporal modeling and cognitive neuroscience so as to augment the synchronization of auditory and motor responses in the dynamic training environment. This study examines the possibilities of intelligent systems in predicting patterns of rhythm and dynamically supporting the user to have a temporal alignment using multimodal feedback. The framework combines data of music beats, motion sensor motions, EEG, and IMU data to record physical and neural entrainment. Preprocessing entails temporal division, beat identification and signal normalization to provide inter-modality consistency. Three predictive architectures are created, namely, Long Short-Term Memory (LSTM), Transformer, and Temporal Convolutional Neural Network (TCNN) to compare their performance in beating timing and synchrony accuracy. The model architecture combines the multimodal-entered information at the initial levels of the model, and uses the modules of temporal prediction, which has the ability to learn to reduce the synchronization time lag by using the self-adaptive feedback mechanisms. As it has been experimentally shown, Transformer-based models are superior to recurrent architectures in terms of their ability to address long-range temporal dependencies, whereas LSTM networks demonstrate resilience to noisy motion data. The discussion brings out the benefits of predictive AI to provide real-time rhythm correction and custom training adaptation. It is used in the field of sports, dance, music pedagogy, and in areas of cognitive rehabilitation, rhythmic accuracy improves motor learning and coordination.