response = openai.audio.speech.create(
model=“tts-1”, # also try tts-1-hd
voice=“alloy”, # other voices are alloy, echo, fable, onyx, nova
input=“Today is a wonderful day to build something people love!”
)
response.stream_to_file(speech_file_path)
getting error :- AttributeError: module ‘openai’ has no attribute ‘audio’
Well, I don’t think other languages will do here. I have tested in Italian, Spanish, Serbian and Greek.
For all but Greek, it sounds like an American actor trying to read these texts in a foreign language, but failing almost miserably. It is not a “slight accent”, as someone reported. The results are unusable - apart from maybe some comic needs.
I believe they are aware of that, as they did not mention languages at any point. It was probably a decision to release this and then work on other languages.
Elevenlabs is great, sounds much better in other languages, but it is around 7x / 15x more expensive for the usage, making it too expensive for regular usage.
I dream of the day they’ll reveal their TTS synthesis process! I’m amazed/horrified when my AI coughs. I’d love to record my own for my particular Assistant implementation.