How Much Audio Do You Need to Clone a Voice?
Updated September 3, 2026
TL;DR: for an instant clone, a short clean sample — think one good minute — is enough. Past that point, extra length adds little; noise, echo and music subtract a lot. Professional cloning across the industry uses far longer datasets for higher fidelity; LeapFun's instant cloning trades some of that fidelity for minutes instead of days.
Quality beats quantity
An instant clone learns timbre: the resonance and consonant shapes that make a voice recognisable. That signal is present in seconds of clean speech. What corrupts it is everything the model can't separate from the voice — room reverb, background music, a fan, compression artefacts from a video call. One clean minute in a closet beats an hour of podcast tape with music underneath.
Recording checklist
- Quiet room, soft surfaces, no echo — a wardrobe, a car in a garage, a duvet over your head if you must.
- One microphone 15–20 cm away, held steady. A phone is fine; a laptop mic across the desk is not.
- Read a paragraph that matches your real use — narration for a narrator, a talk-track for a creator.
- Natural pace, natural energy. The clone inherits the mood of the sample.
- No processing before upload: no noise reduction, no compression, no EQ.
- Record two or three candidates in the same session and audition each.
Instant vs professional cloning
| Instant cloning (LeapFun) | Professional cloning (industry) | |
|---|---|---|
| Sample | A short clean recording | Often 30 minutes to hours of clean studio audio |
| Time | Minutes | Hours to days of training |
| Fidelity | Recognisable timbre; delivery steered by tags and director's note | Closest match, including habits and micro-inflection |
| Best for | Creator channels, brand voices, reading your own book | Flagship narrators, high-end productions |
Signs your sample was the problem
- Thin or nasal timbre → too far from the mic or too much room.
- Hiss or hum → noise in the room or the cable.
- Flat delivery → you read flat; re-record with the energy you want.
- Wrong language colour → record in the language you'll mostly generate; clones read English and Chinese either way.
Try it
Cloning starts on the Starter plan with one voice. Not ready? Preset voices are real samples you can hear now:
Frequently asked questions
Can I clone a voice from a video call recording?
Technically, if it's your voice, but the result will carry the call's compression and room. Record a fresh sample with a phone in a quiet room instead.
Does a longer sample make a better clone?
Only up to a point, and only if it's equally clean. Past a clean minute, the returns are small; a longer noisy sample is worse than a short clean one.
Can I clone from a song or a podcast with music?
Music under the voice degrades the clone. Use a solo speech recording.
Which plan do I need?
Starter ($6 a month) includes one cloned voice; Creator ten; Pro and Team thirty.