ElevenLabs TTS

What is ElevenLabs TTS?#

ElevenLabs TTS turns text into natural-sounding speech, with four model variants on the same endpoint. Send the text and a voice ID and the call works; everything else is optional. The four models trade latency, language coverage, and character cap: Flash v2.5 returns first audio in roughly 75 ms across 32 languages, Turbo v2.5 sits at about 250 ms with stronger quality, Multilingual v2 covers 29 languages at the highest quality with a 10,000-character cap, and V3 reaches 70+ languages with the most dramatic delivery. Output formats span compressed MP3 and Opus, full-rate WAV, raw PCM, and 8 kHz μ-law and a-law for telephony. That coverage is what makes the same endpoint usable for both consumer-facing playback and IVR pipelines.

Key features of ElevenLabs TTS#

Five features cover the surface area you'll touch in a real production pipeline.

Four model variants for the latency-quality-language trade

Flash v2.5 returns first audio in roughly 75 ms across 32 languages, which fits live agents and IVR. Turbo v2.5 sits at about 250 ms with stronger quality. Multilingual v2 is the default for long-form narration. V3 covers 70+ languages with the most dramatic delivery.

Eighteen output formats, including telephony

MP3 and Opus across five bitrates each, raw PCM at five sample rates, full-rate WAV, plus 8 kHz ulaw_8000 and alaw_8000 that drop straight onto Twilio and other voice circuits without an intermediate transcoder.

Voice settings you can actually tune

Stability, similarity boost, style, speed, and a speaker boost toggle. Each has a sensible default, but the controls are precise enough that you can shape the same voice for an audiobook one week and a high-energy promo the next without re-cloning.

Seeded, reproducible output

Pass an integer seed and the same text, voice, model variant, and settings return identical audio. That makes regression tests possible, lets games keep cached lines deterministic, and means a re-render after a copy edit only changes the words you changed.

Live voice catalog including your clones

Cloned voices created inside ElevenLabs show up automatically through the resources catalog, with no redeployment, no SDK update, and no separate fetch path. Pass any returned ID as the voice parameter and the call routes correctly.

Best for#

Variants#

Four model variants share one endpoint. Pick by the latency, quality, and language coverage you need.

Flash v2.5#

First audio in ~75 ms, 32 languages, 40,000 characters per call. Use it for live voice agents, IVR, and any pipeline where end-to-end latency is the headline constraint.

Turbo v2.5#

First audio in ~250 ms, 32 languages, 40,000 characters per call. Same broad language coverage as Flash with stronger quality. Good when you can wait through a quarter-second of extra latency and want a richer voice on the other end.

Multilingual v2#

29 languages, 10,000 characters per call. The default for long-form narration. Pair it with a locked seed and you get consistent voice across chapters, episodes, and series, with no drift between calls.

V3#

70+ languages, 5,000 characters per call. The most expressive of the four, with the biggest dynamic range and the most dramatic delivery. Use it for trailer narration, promo VO, and any cut where the voice has to carry the clip.

Use cases#

Wire a real-time voice agent to Twilio in a day: call the TTS endpoint with output_format set to ulaw_8000, hand the buffer to your media stream, and Flash v2.5's first-audio latency keeps conversation natural. Produce a localized audiobook by writing one chapter generator that loops over locales and swaps between Multilingual v2 (29 languages) and V3 (70+) per market. Build deterministic in-game dialogue by binding each NPC to a fixed voice plus a fixed seed and settings profile. Layer accessible UI playback by pre-rendering common strings server-side and caching the resulting audio URLs.

API examples#

Call ElevenLabs TTS from any language by POSTing to /v1/tasks. Full parameter docs live at docs.unifically.com/models/audio/elevenlabs/text-to-speech.

curl -X POST https://api.unifically.com/v1/tasks \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -d '{
    "model": "elevenlabs/text-to-speech",
    "input": {
      "text": "Hello world, this is a test.",
      "voice": "TX3LPaxmHKxFdv7VOQHJ",
      "model_id": "eleven_flash_v2_5",
      "output_format": "mp3_44100_128"
    }
  }'

Successful submission returns a task_id. Poll GET /v1/tasks/<task_id> or set a callback_url on the request to receive the finished audio URL.