OpenAI lets you describe a voice into existence for GPT-Live
Prompt-to-voice generates a custom voice from a text description, no reference recording or consent clip required, and ships with nine new non-English voices.
OpenAI released prompt-to-voice to all API customers on 16 October. Instead of cloning a voice from a recording, you post a text description of the vocal qualities you want and OpenAI generates a voice to match:
curl --fail-with-body https://api.openai.com/v1/audio/voices \
-H "Authorization: Bearer $OPENAI_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"type": "prompt",
"name": "Support assistant",
"prompt": "A calm, reassuring adult voice...",
"model": "auto"
}'Creating a voice this way is free, and an organisation can hold up to 100 of them. The catch is where you can use one: prompt-generated voices work only with GPT-Live, not the Text to Speech, Realtime or Chat Completions APIs. Speaking through one in a GPT-Live session is billed at the regular session rate. Alongside this, OpenAI added nine built-in voices covering German, French, Hindi, Japanese and Korean.
Why it matters: voice cloning normally means a consent recording and a rights conversation. Describing a voice in text sidesteps both, which makes it far easier to spin up a distinct voice per product, brand or character without recording anyone. It is also a sharp reminder that GPT-Live is where OpenAI is putting its voice investment, not the older APIs.
The caveat: this only helps if GPT-Live is where you are already building. If your voice feature runs on the Realtime or TTS APIs, prompt-to-voice does not reach you yet.