Modeling directly raw waveforms through neural networks for speech processing is gaining more and more attention. Despite its varied success, a question that remains is: what kind of information are such neural networks ...