Ability to train custom Vocabularies
We need to have some way to train custom vocabularies or ontologies for a domain. For example, healthcare or legal phrases that may be commonly spoken in videos.
If you could provide ontology editing capabilities to allow customers to create their own dictionaries or words and phrases and use that to improve the accuracy of speech to text, that would be awesome.
I think the same idea would apply to other processors as well. OCR has some configuration that can improve results. If there was a way to supply a configuration files at an account level, then specify named config files at the API level when uploading a video, you could support more sophisticated application scenarios.