AES Fall Online Convention

While deep learning approaches for beat and downbeat tracking have brought advancements, these approaches continue to rely on hand-crafted, subsampled spectral features as input, restricting the information available to the model. In this work, we propose WaveBeat, an end-to-end approach for joint beat and downbeat tracking operating directly in the time domain. This method forgoes engineered spectral features, and instead produces beat and downbeat predictions directly from the waveform, the first of its kind for this task. Our model utilizes temporal convolutional networks (TCNs) operating on waveforms that achieve a very large receptive field (~30s) at audio sample rates. This is achieved in a memory efficient manner by employing rapidly growing dilation factors, which enable a relatively shallower network architecture. Combined with a straightforward data augmentation strategy, our method outperforms previous state-of-the-art methods on some datasets, while producing comparable results on others.