The multi lanuagage support is very cool but there is not enough control over it. Only for transcription we can pass the language.
Even when prompting specifically for a language, the voice may talk in an accent one time and another time in a perfect native language.
Ideally the language is only changed with a command or tool that sets the language for that entire session. And doesnt start to talk italian when there are some italian words like Pizza.
Instructions that are showcased in OpenAI.fm would be great.
A custom activation word for POS instore solutions would help controlling the start of the voice better and not just start picking up on conversations in the neighbourhood.
Thanks