AI voice generator with a prompt: describe the speaker

An AI voice generator with a prompt creates a new synthetic voice from a written description of the speaker, such as who they are and how they sound, instead of copying a real recording. You then use that voice like any other to speak your scripts. In the Sume app today, Assets → Voices → Generate a voice from text takes a persona description of at least 20 characters: Sume invents a persona, generates a 5-second clip, and clones that voice, and each result is a reusable voi_ voice for text to speech.

Voice design happens in the Sume app and its agent; the public API has no route that creates a voice. The app behavior below is read from the Sume app's current code, and the speech facts from the TTS 1.0 schema in the Sume API reference, the OpenAPI document behind the API reference docs, read on 2026-09-27.

How do I generate a voice from a description in Sume?

• Open Assets → Voices, choose Add a new voice, then Generate a voice from text: “Describe a person. We invent a persona, generate a 5s clip, and clone that voice.”
• Name the voice, then pick its gender and its language.
• Write the persona description. The form lets you submit once it reaches 20 characters.
• In the current code the Assets form creates 3 voices from one description, so you can listen to each and pick the one to use.
• Or ask in a Sume agent chat, for example from the page's Create with agent button. Chat currently creates 1 voice by default, or 3 if you ask for options, and saves the result to Assets → Voices, where teammates can tag it with @ later.

What should the voice prompt describe?

Describe a person, not a use case. The form's own example reads: “A Korean man in his 20s. Usually overflowing with confidence. His voice is strong too — not husky, but bright and cheerful.” It says who the speaker is, what their personality is, and what the voice itself sounds like.

• For an announcer or presenter voice, describe the person who would do that job: their age, their temperament, and how their voice carries.
• The description guides the result but does not guarantee a particular accent or pitch, so listen to the samples before you choose one.
• The language you pick is recorded on the voice. Speaking another language with it later currently makes the API stop for a double-check first.

How do I use the generated voice in text to speech?

Open the voice's More menu, choose Copy ID, and send that id as voice.id on POST /v1/tts-1.0/generate. The API resolves a Voices library id to the stored voice before the job is queued. AI voiceover with your own voice shows the request and the language check.

Delivery is set per request, not stored in the voice: generation_config.speed (0.6–1.5), volume (0.5–2.0), and an optional emotion guide of up to 64 characters, for which the API reference lists no set values.

How is this different from cloning a voice?

Cloning copies a voice from a recording you upload or make; generating from text invents one from a description. Both end up in the same Voices library and work the same way in text to speech. Clone only a voice you have permission to use.

What does speech in a generated voice cost?

It bills like any TTS job: $0.0475 per 1,000 characters, plus a 5.5% agent fee by default, counting spaces and punctuation, for up to 20,000 characters per request. To use several designed voices in one piece of audio, see text to speech with multiple voices.