Extract first and last names from a speech to text transcription?

what is the best instruction prompt format to extract extract first and last names from a speech to text transcription?


Can you post a sample of a transcript? I’ll give it a shot.

will do, but don’t laugh. lol! will post in csv.

ok, see attached as CSV: CSV sample file

and excel: Excel sample file

. we have hundreds of thousands of these examples.

first column is speech to text version. second column is what the first name should be. last column is what the last name should be.

go for it! LOL!

notice in the samples that no matter what instructions you give the callers, they do whatever they want anyways?

Okay so I don’t remember the term for it, but there is too much information loss here. It’s not possible to extract “Angela Aldridge” from “Angela okay” - it’s a language model, not magic.

What you ought to do from here is include large chunks of transcript data so that you can hear someone free associate for a minute.

I’ve been using ASR for my cognitive architecture and GPT3 is pretty good at confabulating the missing or off parts since ASR is messy. You might use the EDIT endpoint to clean it up first.


1 Like

i totally agree with you, that’s why i told you not to laugh. interesting points above, will repost results.