Prompt-to-voice: describe the voice, get the speech
OpenAI's prompt-to-voice turns a text prompt into spoken audio in a single request — plus 9 new GPT-Live voices. Here's what it is, and how it compares to the other direction: unmuss turns your speech into structured text.
The demo works the reverse direction — speak, get clean text back. Prompt-to-voice goes text → speech; unmuss goes speech → text.
What is prompt-to-voice
In October 2026, OpenAI added a prompt-to-voice capability to its GPT-Live service. Instead of picking a voice from a preset list and tuning sliders, you describe what you want in plain text — "a warm narrator, unhurried, slight smile in the voice" — submit your script, and get streaming audio back in one request.
It collapses two old steps into one: voice design and speech synthesis happen in the same call. For developers, that means fewer moving parts between "I need a voice for this character" and hearing it. For everyone else, it's the moment TTS stopped being a menu of presets and started being a directing exercise.
The 9 new GPT-Live voices
Alongside prompt-to-voice, OpenAI added nine natural-sounding voices covering a range of accents, genders, and tonal qualities:
| Voice | Character |
|---|---|
| Nova | Warm, friendly female voice with a subtle American accent. |
| Atlas | Deep, resonant male voice, suited for narration. |
| Luna | Soft, neutral voice — a fit for meditation apps. |
| Ravi | Indian English voice with clear diction. |
| Yara | Middle-Eastern accent voice for regional content. |
| Kai | Crisp, East-Asian English voice. |
| Sofia | European Spanish voice with natural intonation. |
| Mateo | Latin American Spanish voice. |
| Jin | Mandarin Chinese voice with smooth cadence. |
How to use it
-
1. Describe the voice in a prompt — role, accent, pace, energy. The more specific the direction, the closer the result.
-
2. Send your text in the same request — the API streams audio back, so playback can start before generation finishes.
-
3. Iterate on the prompt, not on settings — when a line sounds off, rewrite the direction ("warmer", "slower here") instead of hunting for the right slider.
Prompt-to-voice vs traditional TTS
| Traditional TTS | Prompt-to-voice | |
|---|---|---|
| Voice selection | Pick from a preset list | Describe the voice you want |
| Tuning | Sliders: speed, pitch, pauses | Rewrite the direction in words |
| New character voices | Wait for the vendor to add them | Prompt a new one on the spot |
| Best for | Consistent brand voice at scale | Games, characters, rapid prototyping |
Who it's for — and who it's not for
For
Virtual assistants, interactive games, accessibility tools, and anyone who needs a new voice per character without booking voice actors.
Not for
Going the other direction. If your problem is spoken audio you need as clean text — meetings, interviews, voice memos — that's the speech-to-text side, which is what unmuss does.
Frequently asked questions
- Is prompt-to-voice free?
- It's an API feature, so usage is billed under OpenAI's API pricing — check OpenAI's pricing page for current rates. There's no separate free tier announced for it.
- How is this different from voice cloning?
- Voice cloning copies a specific real person's voice from a sample. Prompt-to-voice designs a new voice from a description — no sample needed, and no consent questions about whose voice it is.
- Can I use it in my own app?
- Yes — that's the point of the API endpoint. One request takes your prompt plus text and streams audio back, which fits virtual assistants, games, and accessibility features.
- What languages does it support?
- The nine launch voices cover American English, Indian English, East-Asian English, European and Latin American Spanish, and Mandarin Chinese, among others. Check OpenAI's docs for the current list.
- How does this relate to unmuss?
- Opposite directions of the same voice pipeline. Prompt-to-voice turns text into speech; unmuss turns your speech into structured, cleaned-up text — try the demo at the top of this page. Full policy: /privacy/.
More voice tools
Work with voice in both directions.
Prompt-to-voice speaks your words. unmuss writes down your thoughts — download it and stop losing ideas to thin air.
Download on the App Store