Venice TTS (/audio/speech)

POST /api/v1/audio/speech converts text to an audio stream or file. OpenAI-compatible — the OpenAI SDK's audio.speech.create() works as a drop-in.

Use when

• You want narration, voice replies, or UI audio from text.
• You need a specific voice family (ElevenLabs, Kokoro, xAI, Qwen 3, Orpheus, Chatterbox, MiniMax, Inworld, Gemini Flash).
• You want streaming audio returned sentence-by-sentence.
• You need style/emotion control on supported models.

For music generation (lyrics + instrumental), see venice-audio-music. For transcription (audio → text), see venice-audio-transcription.

Minimal request

curl https://api.venice.ai/api/v1/audio/speech \
  -H "Authorization: Bearer $VENICE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "tts-xai-v1",
    "voice": "eve",
    "input": "Hello, welcome to Venice Voice.",
    "response_format": "mp3",
    "speed": 1.0,
    "streaming": false
  }' --output hello.mp3

Response is the raw audio (Content-Type matches response_format).

Request schema

Models

Always inspect the entry for your model in GET /models?type=tts — model_spec.voices is the authoritative voice list. Per-model toggles like supportsPromptParam, supportsTemperatureParam, supportsTopPParam live on the internal model definitions but are not currently exposed on /models — treat the request schema below (instructions, temperature, top_p) as the support matrix.

Voice families (by prefix)

• Kokoro — lowercase + language/gender prefix:
• af_*, am_* — American female / male
• bf_*, bm_* — British female / male
• zf_*, zm_* — Chinese
• ff_*, hf_*, hm_*, if_*, im_*, jf_*, jm_*, pf_*, pm_*, ef_*, em_* — French, Hindi, Italian, Japanese, Portuguese, Spanish
• Examples: af_sky, af_bella, am_adam, bm_george, zf_xiaoxiao
• Qwen 3 — Vivian, Serena, Ono_Anna, Sohee, Uncle_Fu, Dylan, Eric, Ryan, Aiden
• xAI — 26 voices. Original five: eve, ara, rex, sal, leo. Flagship multilingual set: altair, atlas, carina, castor, celeste, cosmo, helios, helix, iris, kepler, lumen, luna, lux, naksh, orion, perseus, rigel, sirius, ursa, zagan, zenith
• Orpheus — tara, leah, jess, mia, zoe, dan, zac
• Inworld — Craig, Ashley, Olivia, Sarah, Elizabeth, Priya, Alex, Edward, Theodore, Ronald, Mark, Hades, Luna, Pixie
• Chatterbox — Aurora, Britney, Siobhan, Vicky, Blade, Carl, Cliff, Richard, Rico
• ElevenLabs Turbo — Rachel, Aria, Laura, Charlotte, Alice, Matilda, Jessica, Lily, Roger, Charlie, George, Callum, River, Liam, Will, Chris, Brian, Daniel, Bill
• MiniMax — WiseWoman, FriendlyPerson, InspirationalGirl, CalmWoman, LivelyGirl, LovelyGirl, SweetGirl, ExuberantGirl, DeepVoiceMan, CasualGuy, PatientMan, YoungKnight, DeterminedMan, ImposingManner, ElegantMan
• Gemini 3 Flash — star names: Achernar, Achird, Algenib, Algieba, Alnilam, Aoede, Autonoe, Callirrhoe, Charon, Despina, Enceladus, Erinome, Fenrir, Gacrux, Iapetus, Kore, Laomedeia, Leda, Orus, Pulcherrima, Puck, Rasalgethi, Sadachbia, Sadaltager, Schedar, Sulafat, Umbriel, Vindemiatrix, Zephyr, Zubenelgenubi
• Gradium — the voice selects the language. English: Emma (default), Kent, Eva, Jack. German: Mia, Maximilian. Spanish: Valentina, Sergio. French: Elise, Leo. Portuguese: Alice, Davi

Pass a voice that isn't in the chosen model's list and you get 400.

Voice cloning — POST /audio/voices

Clone a voice from an audio sample and get back a handle (vv_…) you can pass
as voice on /audio/speech. multipart/form-data only.

curl https://api.venice.ai/api/v1/audio/voices \
  -H "Authorization: Bearer $VENICE_API_KEY" \
  -F "model=tts-chatterbox-hd" \
  -F "file=@sample.wav"

A handle is bound to the model that created it. Pass a vv_… handle with a
different model on /audio/speech and the call fails.

Samples in a container outside the per-model allowlist are rejected with 400
before anything is uploaded. Beyond the shared TTS error codes, this endpoint
also returns 403 and 413 (sample too large).

Streaming

{
  "model": "tts-xai-v1",
  "voice": "eve",
  "input": "Hello, this is a long document to narrate. ...",
  "streaming": true,
  "response_format": "mp3"
}

With streaming: true, the HTTP body is a chunked audio stream. Decode as it arrives — useful for latency-sensitive UIs. response_format: pcm pairs well with browser Web Audio API for raw playback.

OpenAI SDK

import OpenAI from 'openai'
import fs from 'node:fs/promises'

const client = new OpenAI({
  apiKey: process.env.VENICE_API_KEY,
  baseURL: 'https://api.venice.ai/api/v1',
})

const mp3 = await client.audio.speech.create({
  model: 'tts-xai-v1',
  voice: 'eve',
  input: 'Hello from Venice.',
  response_format: 'mp3',
})

await fs.writeFile('hello.mp3', Buffer.from(await mp3.arrayBuffer()))

Emotion / style (Qwen 3 only)

{
  "model": "tts-qwen3-1-7b",
  "voice": "Vivian",
  "input": "We did it!",
  "prompt": "Excited and energetic.",
  "temperature": 0.9,
  "top_p": 0.95
}

For other families, emotion comes from the voice choice itself (e.g. Inworld Hades vs Pixie). prompt / temperature / top_p are silently ignored.

Errors

Gotchas

• input hard cap is 4096 chars. For books / long content, split on sentence boundaries and concatenate audio client-side.
• streaming: true + SDKs: some OpenAI SDK versions don't expose streaming for audio.speech.create; call the REST endpoint directly and consume the HTTP body.
• speed compounds with model internal speech rate — extreme values (0.25, 4.0) often sound unnatural; keep within 0.8–1.3 for narration.
• Voice names are case-sensitive (eve ≠ EVE, af_sky ≠ AF_SKY).
• Cloned voices expire. Chatterbox handles die 7 days after creation no matter what; MiniMax handles die after 7 days of no use. Re-clone rather than assuming a handle you stored last month still resolves.
• Gradium has no language parameter. Pick the voice for the language you want.