Training The Transcription

I have a question about training Deepgram or alternatives for accuracy when it comes to addresses. I want to load up all the suburbs for the area our client is working in so that it knows what to pull from. I know Deepgram you can add 100 keywords... But ideally, I want more than that. I want 500 or so. And I'm cautious that adding even 100 will cause latency (Plus how do I know which are the top 100 to add). We are already verifying the address via Google but if the transcription falls over it can be a 2 or 3 attempt to get some more local pronunciations across the line. Any thoughts would be appreciated.

2 comments