Vapi is the Voice AI platform for developers.

Vapi AI

I've built an AI voice assistant to help me research restaurants while I'm driving. I've tried different voices and different ways of outputting things from my tool, but I can't get the voices to pronounce the restaurant names. They will say literally everything else in the output, and the logs show that the model is outputting the restaurant name to the voice, but the voice will not pronounce the restaurant names like 98% of the time. What can I do? I've tried vapi voices, elevenlabs voices, and cartesia voices.

Here's one call where this happened:
call-id: e03ee0c8-09d6-4c9d-a43a-11607271e549
timestamp: 21:05:58:375
response expected: the name of the restaurant: Tavernetta to be pronounced.
response received: the voice skipped the restaurant name.

I seem to have improved it by adjusting the system prompt on vapi to emphasize writing in full sentences for a script that will be read aloud. Previously it was just listing restaurant names, addresses, ratings, and descriptions like you might find online and that format seems to not work well with voice.

there was a pull request which made some changes to the voice formatting. please test it and let us know if there are still some issues with it