How to use Inference API to perform speech recognition

Beginners
2

The default seems to be up to 2 MB. However, model authors and Spaces authors have the means to increase the limit…?
I also heard that the limit is relaxed when using Inference Endpoint.