Closed jmtatsch closed 1 month ago
It doesn't really die, this was premature...
it should error back to the user instead of this behavior. I'll rename & reopen the bug for tracking.
Thanks for the report.
Yes, especially with the /v1/audio/speech endpoint. If you pass in a model other than "tts-1-hd" or "tts-1" you get:
server-1 | File "/app/speech.py", line 193, in generate_speech
server-1 | ffmpeg_args.extend(["-"])
Good point, thanks for the feedback!
Here, I had by chance the non existing voice 'alls'