The standard way of training deep learning systems for speech recognition, OCR, image recognition, and natural language understanding nowadays is end to end, with the system learning its own features. Previously, a good front end was necessary to pre-process signals before training the system.