OpenAI lets developers describe the voice their app should have

A voice brief, then a saved voice

OpenAI’s custom-voice guide lets a developer specify qualities such as accent, pitch, texture and pace. A support assistant might need a calm delivery; a guide might need more energy. The prompt produces a saved voice that can be reused in later sessions, without supplying a recording of a person.

The compatibility boundary matters: prompt-generated voices work with GPT-Live, but not the Text to Speech, Realtime or audio Chat Completions APIs. OpenAI’s announcement demonstrates how to make and use a voice.

French and German join the preset list

The 16 October release also adds nine GPT-Live presets: three German voices, two French, one Hindi, two Japanese and one Korean. OpenAI says they were created with native-speaking actors. For a team building a French- or German-language service, that provides additional ready-made voices alongside the new design option.

Free to create, paid to speak

Creating a voice costs nothing, but generating speech—including listening in the Playground—uses regular GPT-Live session billing. There is an organisation-wide ceiling of 100 prompt-generated voices, including voices an app creates for its customers.

For a brand, custom also does not mean exclusive: OpenAI warns that other users may generate similar voices. Its rules require adult voices, prohibit impersonation and require listeners to be told the speech is AI-generated. The useful promise is more control over vocal character. Independent tests of consistency across real conversations remain a follow-up question.