ElevenLabs Voice Clone v4
Clone a voice from an MP3 or WAV sample, then synthesize up to 10,000 characters of speech using ElevenLabs Eleven v4 or v4 Turbo.
## Overview
Voice Clone V4 clones a speaker from a short recording, then reads your script in that same voice. It uses ElevenLabs Instant Voice Cloning plus Eleven v4 or Eleven v4 Turbo speech synthesis. You get an MP3 voiceover you can drop into videos, apps, or narration workflows without hiring the original speaker again.
## What you can build
- Personalized narration for product demos, tutorials, and walkthrough videos
- Character dialogue for games and interactive stories using a consistent speaker identity
- Branded voiceovers for ads and social clips, with repeatable tone and cadence
- Multilingual reads in the same voice, while keeping the voice identity consistent
- Dataset generation for QA, call flows, or agent scripts that need natural speech
## Inputs
- A reference voice recording you want to clone, provided as an MP3 or WAV file. Keep it to a single speaker, with no music or background noise, and within 3 minutes and 10 MB.
- The script you want spoken, provided as plain text up to 10,000 characters. You can include bracketed performance tags like [whispering], [laughing], [shouting], or [excited] inside the text.
- A choice of speech model variant: Eleven v4 for produced content quality, or Eleven v4 Turbo for faster generation at half the price.
- The MP3 output quality you want. Pick a sample rate (22.05 kHz or 44.1 kHz) and a bitrate between 32 kbps and 128 kbps.
- An optional language selection. You can let the model detect the language from your text, or force a specific language so numbers and dates are read correctly.
- An optional switch to return timing data alongside the audio. Use this when you need word timing for captions, highlighting, or alignment.
- An optional text normalization preference. This controls whether numbers, dates, and abbreviations are expanded into spoken words before synthesis.
- Optional delivery controls on a 0 to 1 scale that trade off expressive variation versus consistency, and voice similarity versus naturalness.
- An optional reproducibility seed, provided as an integer from 0 to 4,294,967,295, to make repeated runs sound more alike.
## Outputs
- An MP3 file containing the generated speech in the cloned voice, encoded at the sample rate and bitrate you selected.
- If you request timing data, an additional JSON file with the start and end time of every word, suitable for captioning or word highlighting.
## Recommended settings
- For most general narration: keep the delivery controls at their defaults, 0.5 for stability and 0.75 for similarity.
- For the closest clone match: use 1 to 2 minutes of clean reference audio with one speaker and no reverb.
- For web and app playback: use 44.1 kHz MP3 at 128 kbps.
- If your script includes many numbers and dates: keep text normalization on automatic unless you have custom spellings.
## Limitations
- Clone quality depends heavily on the reference recording. Background noise, room echo, multiple speakers, music, and heavy compression reduce similarity.
- Very short samples of only a few seconds give a weaker match than a minute or more of clean speech.
- Instant voice clones can struggle with very unusual accents or highly stylized voices.
- Output isn't perfectly deterministic. A fixed seed reduces variation, but it can't guarantee identical audio.
- Messy text inputs can degrade speech. Emojis, uncommon symbols, and unformatted copy-pastes can cause misreads or unstable pronunciation.
## Safety & compliance
- Only clone voices you own or have explicit permission to use.
- Samples that match a protected voice are rejected and the task stops with a message.
- Generated audio may be detectable as AI-generated through classifier tooling. Use clear disclosure where your policies or laws require it.
Example prompts
Great starting points for voice clone v4.
API quick start
Run voice clone v4 with a single API call.
{
"prompt": "A pinch of salt, a splash of olive oil, a…",
"inputAudio": "https://your-cdn.com/input.mp3",
"model": "eleven_v4_turbo",
"outputFormat": "mp3_44100_128"
}curl -X POST "https://api.wiro.ai/v1/Run/elevenlabs/voice-clone-v4" \
-H "Content-Type: application/json" \
-H "x-api-key: YOUR_WIRO_API_KEY" \
--data-binary @- <<'JSON'
{
"prompt": "A pinch of salt, a splash of olive oil, a…",
"inputAudio": "https://your-cdn.com/input.mp3",
"model": "eleven_v4_turbo",
"outputFormat": "mp3_44100_128"
}
JSONOther models you might like
Explore similar audio & speech models.



