Text to speech is live — 19 voice models from $0.0014 per 1,000 characters

What's new

Nineteen text-to-speech models, spanning three useful tiers. At the top, MAI-Voice-2, MiniMax Speech 2.8 HD and Gemini 3.1 Flash TTS for production narration where quality is the constraint. In the middle, MAI-Voice-2 Flash, Aura-2, Voxtral, Grok Voice, Qwen TTS and the Fish Audio family for voice agents and everyday synthesis. At the bottom, Kokoro, Orpheus, Sesame CSM and Zonos — cheap enough that narrating your entire content library stops being a budget decision. Two capabilities stand out: Fish Audio S2.1 Pro clones a voice from a single sample attached to the request, and Voxtral turns emotion into a voice ID so the same line reads angry or cheerful without prompt engineering.

How to call them

Nothing changes in how you integrate. Submit the text to `/v1/media/queue`, then poll `/v1/media/status/{jobId}` until it reports `completed` and read the base64 audio out of `result.audio`. Pick a voice with `options.voice` — every model has a default if you omit it — and MP3 comes back unless you ask for `pcm`.

curl -X POST https://api.yepapi.com/v1/media/queue \
  -H "x-api-key: $YEPAPI_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "microsoft/mai-voice-2",
    "prompt": "Your report finished processing overnight.",
    "options": { "voice": "en-US-Harper:MAI-Voice-2" }
  }'

Voice cloning without an enrolment step

Fish Audio S2.1 Pro supports stateless cloning: attach a short reference sample as `audioData` and the generated speech mimics that voice. There is no voice to create, upload, name, or delete beforehand — the sample travels with the request and nothing is retained. Add `options.referenceText` with the sample's transcript to improve fidelity.

curl -X POST https://api.yepapi.com/v1/media/queue \
  -H "x-api-key: $YEPAPI_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "fish-audio/s2.1-pro",
    "prompt": "This line is spoken in the cloned voice.",
    "audioData": { "mimeType": "audio/mpeg", "base64": "..." },
    "options": { "referenceText": "I used to rule the world." }
  }'

Pricing

Speech is billed on the length of the text you send, in UTF-8 bytes — for English, one byte is one character. That means the cost is known before synthesis starts, so the balance check at submit time quotes the exact final charge rather than an estimate. Rates run from $0.0014 per 1,000 characters (Kokoro 82M) to $0.2110 (MiniMax Speech 2.8 HD), with a $0.01 minimum per job. It comes out of the same prepaid balance as your text, image, video, SEO and scraping calls — no subscriptions, no character bundles, no minimum commitment.