Training dynamic models using early exits for automatic speech recognition on resource-constrained devices

Re-assign community