Test Spanish/non-English language performance

[nonprofittechy]

1. Do we need to set up a language detection prompt at the beginning?
2. Will some of the models detect automatically, and can we skip this?
3. Do we want to have a dedicated Twilio flow to direct to a different endpoint based on language?

[jpagh]

If just prompting then we need to test how it handles other languages, especially with regard to function calling

[jpagh]

In be07bba I added a lot of prompting that has proven to be effective in my Spanish language automated test. I don't speak Spanish, so I can't promise that it's the best, and the Text-to-speech models sound super white, but I think it will work.

We MUST test this with whatever STT/TTS we're thinking of changing to in case there are issues I can't foresee. (since I'm not in love with Gemini/OpenAI's STT/TTS).

I did some other work to constrain some of the function inputs better (like date of birth) which should help all languages. I also added very basic transliteration so that unicode characters are converted to basic ASCII characters in the state and thus for LegalServer.

[jpagh]

I need to ask Leslie about Spanish.

Deepgram Flux doesn't have Spanish support. If they want Spanish, and we decide we want Deepgram Flux, maybe we could use a Dialpad IVR to choose the language and then have a bot (or toggle) for each (only changing STT/TTS/VAD/Turn).

[jpagh]

From Leslie:

Spanish would be great.  We get some non-English speakers, mostly Spanish, but they are very often not in our service area.  We don't get many other languages other than Spanish and when we do, they are almost always out of our service area.