Checked 17 October 2026. One material, previously uncovered announcement met Tecrave’s 48-hour threshold.
OpenAI releases prompt-generated custom voices for GPT-Live
OpenAI added prompt-to-voice to its API on 16 October. All API customers can now describe an adult voice in plain language, create it through the API or Voice Playground, save it and reuse it in GPT-Live sessions. The release also adds nine natural voices across German, French, Hindi, Japanese and Korean.
Creating a prompt-generated voice is free. Listening to it or using it in production is billed at the normal GPT-Live session rate, which is charged by session duration. OpenAI’s documentation says an organisation can save up to 100 prompt-generated voices.
Why it may matter: a small business with a developer or integration partner can prototype a consistent voice for a telephone assistant, booking service or multilingual support experience without recording a voice actor first. A practical test would be to create one support voice, run it against common customer questions, then assess clarity, pronunciation, interruption handling and escalation to a person before wider use.
Analysis: this is useful mainly for businesses already building with the OpenAI API. It is not a general voice-over tool. Prompt-generated voices currently work only with GPT-Live, not the Text to Speech, Realtime or Chat Completions audio APIs. OpenAI also requires developers to tell listeners that speech is AI-generated, prohibits children’s voices and impersonation, and warns that generated voices are not guaranteed to be unique. Those restrictions make disclosure, brand-safety review and human fallback part of the implementation, not optional extras.
Primary sources: OpenAI API changelog, custom voices documentation and GPT-Live guide.
Prepared with AI assistance from linked primary sources.