AI voice clone: sing with your own voice
A voice clone is a reusable asset built from a short recording of you. Once it exists you can sing new songs or cover existing ones in that voice - creating it is free, and each generation with it returns two tracks.
Voice creation free · 1.2 CNY per generation for 2 tracks · unlimited downloads
What the sample needs
- 10 seconds minimum - 20 to 60 seconds of continuous speech or singing is much more stable.
- Clean: no backing music, no reverb, no fan or street noise in the background.
- Natural: talk or sing normally; reading single words or shouting produces a worse voice.
- Under 10 MB, in mp3, wav or m4a.
- Yours, or licensed - a voice is reusable, so using someone else's without permission is a legal problem, not just a policy one.
The flow
Why results disappoint, and what to change
| Symptom | Likely cause | Fix |
|---|---|---|
| Muffled, chorus-like vocal | Backing music inside the sample | Re-record an a cappella take |
| Unstable pitch and timbre | Sample too short | Use 20+ seconds and re-create the voice |
| Background hiss in the output | Noisy room | Record somewhere quiet, re-create |
| Only works in one style | Sample is stylistically narrow | Include both speech and sung lines |
What a voice profile is actually good for
| Use case | Why a voice profile helps |
|---|---|
| Custom songs as gifts or commissions | The client hears their own voice on the track instead of a generic one |
| Demos before studio time | Sketch the arrangement and the vocal line without booking a session |
| Voiceover-adjacent music beds | Sing short hooks and tags that match an existing artist identity you own |
| Accessibility and personal projects | Produce sung material when recording yourself is not practical |
The constraint is the same in every row: it has to be a voice you own or have permission to use. A voice profile is a reusable asset, so the rights question does not go away after the first song.
Voice profile or voice clone? Same thing
Different products use different names - voice clone, voice profile, custom voice, persona. Mechanically they are the same idea: encode a speaker from a sample, then use that encoding when generating. What changes between platforms is how the sample is validated (some require a second recording as verification) and how long the voice stays usable.
On our side the voice is free to create and 1.2 CNY per generation to use, and the sample requirements above are the only hard rules. If a platform asks for a verification recording, that is an anti-abuse check, not a quality check.
FAQ
How much audio do I need to clone a voice?
Ten seconds is the minimum. Twenty to sixty seconds of clean, continuous speech or singing is noticeably more stable, and the file should stay under 10 MB.
Does creating a voice cost anything?
Creating the voice is free. Generating a song with it costs 1.2 CNY per run and returns two tracks.
Why does the result not sound like me?
Usually the sample: backing music, room noise or a very short clip. Try a quiet room, 20-60 seconds of continuous speech plus a little singing, and no reverb.
Do voice profiles expire?
They can become unavailable over time, so check that the voice is usable before a batch run rather than treating it as a permanent asset.
Make a voice, then use it
Sign-up takes an email address; the voice itself is free to create.