Hi, I am trying to make audio transcriptions that translate to English using openai.audio.translation.create per:https://platform.openai.com/docs/guides/speech-to-text.
It always transcribes, but whether or not it translates to English seems to be random. I could only get this endpoint to produce transcriptions in the source language unless I set the temperature to 1. Once I did, the endpoint would transcribe (testing with Spanish or Japanese) to English about once every 4-5 times for the my Spanish test and once every 2-3 times for my Japanese test.
OpenAI’s provided translation example does not use the temperature setting at all. I also tried adding a prompt but that does not seem to help.
Any ideas?