GPT-Live custom voices: prompt-to-voice API and international presets
Source
developers.openai.com
Author
OpenAI DevelopersTop Viber
Date
Why it matters
You can create a saved GPT-Live voice from a text description with one API call and no recording or consent step. Prompt voices work only with GPT-Live, and recording-based custom voices need enabled access.
Key takeaways · AI-distilled
Creating a prompt-generated voice is free and capped at 100 per organization, a limit that also covers voices an app creates for its customers; speech with the voice in GPT-Live bills at regular session rates.
The create call returns only metadata, with no audio or preview; OpenAI advises retaining the original description and including it in session.instructions alongside the saved voice ID for voice performance.
Voice creation through /v1/audio/voices, including from recordings, is not compatible with Zero Data Retention, although GPT-Live speech generation itself is ZDR-eligible when enabled for the organization or project.
Recording-based voices stay limited to eligible customers: at most 20 per organization, a scripted consent phrase in one of 16 languages, and a same-speaker sample of up to 30 seconds with at least five seconds of actual speech.
gpt-live-1 supports custom voices with English accents set through session instructions such as "Speak British English", and the voice is fixed at session start, so trying a different voice requires a new session.
