Peter Salas

Yeah, it can definitely be problem, especially when multiple fields are being showed and it's embarrassingly bad when it's `a.really.long.sequence.of.objects.ending.in.fieldA` and `a.really.long.sequence.of.objects.ending.in.fieldB`. The ideas around trimming sound good to me...maybe...

Oops, hit the wrong button :)

Thanks for the additional details! It looks like React Native doesn't have built-in support for `TextEncoder`. Can you try the workaround suggested in this [thread](https://github.com/hapijs/joi/issues/2141) (namely, using a polyfill)?

> @petersalas did you had a chance to do performance comparison of non-chunked vs chunked prompt execution with your changes? Not rigorously -- I did some ad-hoc runs of `offline_inference_audio_language.py`...

FWIW I ran into this issue with latest py-spy, saw similar segments as are described above, and confirmed that @yinghai's branch worked perfectly.

@duryabaziz Sorry for the delay -- I'd be happy to help in our [Discord](https://discord.com/invite/Qw6KHxv8YB)! Please join and start a thread in #ultravox-realtime-help.

This repo is for the Ultravox model itself -- for questions or help with the Ultravox Realtime service, please use our Discord, where we're happy to help with prompting techniques...

@DarkLight1337 can you elaborate a bit? I pulled the branch down but the audio tests seem to pass. Is there a call stack with the failure I can look at?