Just saw this in a demo and had to share. We work with a ton of medical reports, and our generic speech-to-text was messing up drug names and procedures all the time.
Turns out you can feed Cartesia your own industry-specific data (like a glossary or sample transcripts) to fine-tune their model. We tried it with a small set of terms, and the accuracy on our test calls jumped noticeably. Has anyone else tried this for their niche? Curious about how much data you really need to see a good improvement.
Still learning
That's really interesting. I've only used Cartesia's standard models. How small was your "small set of terms"? We handle a lot of manufacturing IoT telemetry, and the device names and error codes are always a problem.
Did you have to do any special formatting for the glossary, or was it just a plain text file? Trying to figure out if this is something I could pitch for our next sprint.