Low-Level Physiological Implications of End-to-End Learning of Speech Recognition
This work offers incremental insights into biological hearing for researchers in speech processing and neuroscience.
The study investigated whether end-to-end speech recognition systems can provide insights into human hearing mechanisms, finding that they learn both narrow and wide-band filters, with the latter likely arising in the mid-brain rather than the cochlea.
Current speech recognition architectures perform very well from the point of view of machine learning, hence user interaction. This suggests that they are emulating the human biological system well. We investigate whether the inference can be inverted to provide insights into that biological system; in particular the hearing mechanism. Using SincNet, we confirm that end-to-end systems do learn well known filterbank structures. However, we also show that wider band-width filters are important in the learned structure. Whilst some benefits can be gained by initialising both narrow and wide-band filters, physiological constraints suggest that such filters arise in mid-brain rather than the cochlea. We show that standard machine learning architectures must be modified to allow this process to be emulated neurally.