Quick Answer
The simplest free way to clone your voice is a hosted tool with free credits: Percify's free plan covers a voice clone (5 credits, from 15 to 30 seconds of speech), with commercial rights on every plan, while ElevenLabs puts Instant Voice Cloning on its $6 Starter plan. For free and unlimited, run an open model yourself: OpenVoice, Chatterbox and Zonos allow commercial use; F5 TTS's models are non commercial, and XTTS v2 comes under Coqui's own model licence.
How to clone your voice free with AI, checked on 15 September 2026: which hosted tools include cloning free, how long a sample you need, and which open models allow commercial use.
There are two free ways to clone your voice. A hosted tool runs in the browser and gives you free credits; the catch is usually that cloning, or commercial use, starts on a paid plan. An open model runs on your own computer with no limit; the catch is a graphics card, some setup, and a licence that may not allow commercial use.
Either way, the recording decides most of the quality: a clean room, one voice, and sentences with some range. Every plan, price and licence below was read on the maker's own page on 15 September 2026.
Free voice cloning options at a glance
| Option | Free? | Clone from | Commercial use | Paid from |
|---|---|---|---|---|
| Percify | Yes: the free plan's 10 credits cover a clone (5 credits, once) | 15 to 30 seconds of speech, at least 5.6 | Every plan, the free one included | Starter $6.99 a month |
| ElevenLabs ↗ | Free plan, but no voice cloning | Instant Voice Cloning from Starter; Professional from Creator | Commercial licence from Starter | Starter $6 a month |
| Descript ↗ | Free plan; custom voice clones listed from Hobbyist | Not stated on the pricing page | Watermark free exports from Hobbyist | Hobbyist $16 a month billed yearly, or $24 |
| OpenVoice ↗ | Free, open source (MIT) | A reference clip | Yes: "free for commercial use" | Your own GPU |
| Chatterbox ↗ | Free, open source (MIT) | A reference clip; Multilingual V3 covers 23+ languages | Yes, under MIT | Your own GPU |
| Zonos v0.1 ↗ | Free, open weights (Apache 2.0) | A 10 to 30 second sample | Yes, under Apache 2.0 | Your own GPU |
| F5 TTS ↗ | Free code (MIT) | A reference clip | No: the pretrained models are CC BY NC | Your own GPU |
| XTTS v2 ↗ | Free code (MPL 2.0) | A reference clip | The model is under the Coqui Public Model License: read its terms first | Your own GPU |
How to clone your voice on Percify, step by step
- Record 15 to 30 seconds. A soft room, a steady distance from the microphone, and sentences with range: a question, a statement, a number. Percify needs at least 5.6 seconds of clear speech; shorter samples are padded by repetition, which makes the clone flatter.
- Create the voice. Clone Yourself makes it for 5 credits, charged once, and it is then yours to reuse. The free plan's 10 credits cover it, with no card, and cover a face from one photo too.
- Make it speak. Type any script and your voice reads it, in a talking video of you or on its own. In the playground, Zonos 2 ↗ clones a voice from a short sample and costs 2 credits per minute of speech.
- Listen for the giveaways. Test with a script unlike your sample, with numbers and names; that is where a weak clone slips.
For a talking video of yourself in that voice, see making a realistic AI avatar of yourself; for short videos, see Reels with an AI avatar.
Running an open model yourself
Open models are free with no credits and no caps, but you install them and run them on a graphics card. Read the licence before you publish anything:
- OpenVoice ↗, from MIT and MyShell, is MIT licensed since April 2024, and its authors call it "free for commercial use".
- Chatterbox ↗, from Resemble AI, is MIT licensed; its Multilingual V3 model covers 23+ languages and Turbo is a smaller English model.
- Zonos v0.1 ↗, from Zyphra, has open weights under Apache 2.0 and clones from a 10 to 30 second sample; on an RTX 4090 it makes about 2 seconds of audio per second of compute.
- F5 TTS ↗ releases its code under MIT, but its pretrained models are CC BY NC because of their training data, so no commercial use.
- XTTS v2 ↗ comes from Coqui; the code is MPL 2.0, but the model is under the Coqui Public Model License rather than an open source licence, so read its terms before any commercial use.
What makes a clone sound like you
- One speaker, no music. Background sound and a second voice end up in the clone.
- Range, not a monotone read. A question, a statement and a number teach the model more than a long flat paragraph.
- The same microphone you will publish with, so the clone and your real recordings match.
- A test script unlike the sample. Names, numbers and questions show a weak clone fastest.
Only your own voice, or with permission
Clone only voices you have the right to use. Your own is the easy case; a colleague, a client or a public figure needs their permission, whatever a tool accepts. Check each tool's terms, and each open model's licence, before a clone goes into anything you publish or sell.
Ready to Create Your Own AI Avatar?
Make a talking video of yourself from one photo, in your own voice. No watermark on any plan.
Get Started FreeFree to start · no credit card
Got questions?
Frequently asked
Yes. On 15 September 2026, Percify's free plan covered a voice clone, 5 credits from 15 to 30 seconds of speech, with commercial rights; ElevenLabs' free plan did not include cloning, which starts on its $6 Starter plan. Open models such as OpenVoice, Chatterbox and Zonos are free with no limit if you run them on your own computer.
Percify asks for 15 to 30 seconds with some range, and at least 5.6 seconds of clear speech; shorter samples are padded by repetition and sound flatter. Zonos v0.1 clones from a 10 to 30 second sample.
On 15 September 2026: Percify on every plan, the free one included; OpenVoice and Chatterbox under MIT; Zonos v0.1 under Apache 2.0. F5 TTS's pretrained models are CC BY NC, so non commercial, and XTTS v2's model is under the Coqui Public Model License, whose terms you should read first. ElevenLabs' commercial licence starts on Starter.
Not on its free plan, as of 15 September 2026. Instant Voice Cloning and a commercial licence start on Starter at $6 a month, and Professional Voice Cloning on Creator.
Cloning your own voice is the easy case. For anyone else, a colleague, a client or a public figure, you need their permission, whatever a tool accepts, and you should check each tool's terms and each model's licence before publishing.
