In the depths of Siri, it's churning out confidence scores for its dictation. I've accessed this on ios (via simulator on my mac), by tinkering with a package that connects Siri with react native.
It returns something like
{"value": [[{"confidence": .98, "substring": "just"} , {"confidence": .98, "substring": "a"} ,{ "confidence": .80, "substring": "test"} ]]}
Now I have a bunch of files in my mac that I want transcribed -- with confidence scores, as above. It would really be best to do this with Siri.
It seems like my options are
make a little RN app that pulls in audio files from my mac, plays them, transcribes them, and saves the results to csvs (but how can I get the iOS simulator to interact with my files??)
tinker with the python package that accesses macbook's speech to text capabilities -- which theoretically seems simpler, but I can't understand how it works
give up and use Google's API (I'd rather not, for various reasons)
something else???
How would you tackle this? Has anyone done anything like this?
