Instant clone · no training · 10 languages
Upload or record a short clip of a voice you have permission to use, type what it should say, and get it back in that voice — even in another language. No training wait: the clone is made fresh for every clip.
Updated by the Kenerate AI team
On this page
Quick answer
Updated
An AI voice clone copies a voice from a short sample and reads any new text in it. On Kenerate AI you upload or record a 3–15 second sample, type a script of up to 5,000 characters, and the Qwen3 instant clone speaks it in 10 languages, with no training. Only clone voices you own or have permission to use.
Listen first
Each sample on the left is an AI voice (so nobody's real voice was copied). The clone on the right reads a line the sample never said.
Sample · 7 s
Callum · English
My name isn't important. What matters is the story, and tonight I'm going to tell you how the old mill burned down.
ElevenLabs Multilingual v2 · 0:07
Clone
Instant clone · English
The fire started in the grain room just after midnight. By the time the bells rang, the whole east wall was glowing orange.
Qwen3 Voice Clone · 0:07
A 7-second gravelly sample becomes a storyteller reading a new paragraph. Timbre and grit carry over; pacing follows the new text.
Sample · 6 s
Upbeat Woman · English
Hi! I'm recording this in my kitchen, so excuse the echo. I just want to see if this actually sounds like me.
MiniMax Speech 2.8 HD · 0:06
Clone
Instant clone · English
Okay, that's honestly a little spooky. It even does the thing where I speed up when I get excited.
Qwen3 Voice Clone · 0:04
A casual 6-second “kitchen recording” — the clone keeps the bright, quick delivery.
Sample · 6 s
Upbeat Woman · English
Hi! I'm recording this in my kitchen, so excuse the echo. I just want to see if this actually sounds like me.
MiniMax Speech 2.8 HD · 0:06
Clone
Instant clone · Spanish
Y ahora en español: es la misma voz, pero hablando un idioma que nunca grabé.
Qwen3 Voice Clone · 0:05
Same English sample, new language: the clone speaks Spanish in the same voice. Qwen3 clones speak 10 languages.
More voices
No clip to clone? Describe the voice instead. These two voices were made from the text descriptions shown.
Qwen3 Voice Design · Alibaba Qwen
Voice descriptionA warm, slightly husky woman in her forties with a gentle Southern American drawl, speaking slowly like a country radio host at dawn
Mornin', y'all. The coffee's on, the fields are quiet, and we've got three hours of the best old songs you ever heard.
Qwen3 Voice Design · Alibaba Qwen
Voice descriptionA low, husky female voice with a slight rasp, calm and confident, like a late-night jazz radio host
It's two a.m., the city's finally quiet, and this next record goes out to everyone still awake.
How to
Step 1
Upload an audio file, record straight from your microphone, or pick a clip from your gallery. Use a voice you own or have permission to clone.
Step 2
A matching transcript usually gives a closer clone.
Step 3
Up to 5,000 characters, in any of 10 languages.
Step 4
The clone reads your script; download the MP3. To reuse the voice, save it as a MiniMax clone.
See it in the app
Open Voice Studio in Text to speech and type the words you want spoken, or start from a script template.
Text to speech, emotions & tags, Hindi, voice design, cloning & chat
Chapters
Use cases
With permission, a clone saves re-recording every time the script changes.
Fix a flubbed line or add a new intro without setting up the mic again.
Record in English, publish a Spanish, German or Japanese version in the same voice.
Patch one sentence in a lesson instead of re-recording the whole module.
Keep greetings in your own voice while changing the words anytime.
Clone a consenting actor's session so late script changes don't need a new booking.
People who may lose their voice can keep a clone of it while they still have recordings.
Engines
Instant clones run on Qwen3; saved clones run on MiniMax Speech. Facts from the Voice Studio's code.
Alibaba Qwen
Instant clone from a 3–15 s sample, no training
Open Qwen3 Voice Clone
MiniMax
Most natural MiniMax voice; 7 emotions, (laughs)/(sighs) interjections, pronunciation dictionary
Open MiniMax Speech 2.8 HD
MiniMax
Near-HD quality, faster; same emotions and interjections
Open MiniMax Speech 2.8 Turbo
Alibaba Qwen
Describe a voice in words and hear it
Open Qwen3 Voice DesignControls
The sample matters more than any setting. These are the knobs the studio gives you.
About 10 seconds of continuous speech is the sweet spot: long enough for timbre, short enough to stay clean.
One speaker, a quiet room, no background music or effects. Noise gets cloned too.
Typing exactly what the sample says helps the model line sounds up with words — usually a closer match.
The clone can read English, Chinese, Japanese, Korean, German, French, Spanish, Italian, Portuguese or Russian.
Save a clone once (with optional noise reduction) and use it with emotions, speed and pitch on every MiniMax Speech engine.
Per clip for instant clones; split longer scripts into sections.
Scripts
Mix questions, numbers and emotion to hear how well the clone holds up.
Range test
Wait, really? That's amazing! Okay, okay — let me read the numbers again: forty-two, one hundred and seven, and nine thousand.
Calm narration
The morning was quiet. Somewhere down the road a dog barked once, then thought better of it.
Energetic
Let's go! Big news today, and trust me, you're going to want to hear this one.
Cross-language
Bonjour à tous ! Aujourd'hui, je vous présente mon nouveau projet.
Tips
Most bad clones come from bad samples.
Record at normal speaking distance in a quiet room — soft furnishings beat bare walls.
Speak in the style you want back: a calm sample gives a calm clone, an excited one gives excitement.
Avoid music, laughter from others and long silences in the sample.
Trim the sample to clean speech only; 8–12 seconds is plenty.
Add the reference text when you can.
Only clone your own voice or a voice whose owner has agreed — and never use a clone to impersonate someone.
Example uses
Illustrative examples of typical workflows, not customer reviews.
A creator can fix a flubbed line or add a new intro in their own cloned voice without setting up the mic again.
A teacher can patch one sentence in a lesson instead of re-recording the whole module.
Someone who may lose their voice can keep a clone of it from recordings they still have.
FAQ
9 questions
Yes — a free Kenerate account starts with free credits, enough to try instant voice cloning on your own sample. There's no watermark on the audio, and credits never expire.
Link to this answer3 to 15 seconds. About 10 seconds of clean, continuous speech from one speaker gives the best results.
Link to this answerThere's no training step. The Qwen3 instant clone is created when you generate, so you hear the result as soon as the clip is ready.
Link to this answerYes. Instant clones can read English, Chinese, Japanese, Korean, German, French, Spanish, Italian, Portuguese and Russian — even if the sample was in another language.
Link to this answerAn instant clone (Qwen3) is made fresh from the sample for each clip. A saved clone (MiniMax) is created once, stored in your account, and then works with MiniMax Speech engines, including emotions, speed and pitch.
Link to this answerNo. Only clone your own voice or a voice whose owner has given permission. Cloning someone without consent, or using a clone to impersonate or deceive, isn't allowed.
Link to this answerYes. The studio can record from your microphone, or you can upload a file or pick audio from your gallery.
Link to this answerYes, if you have the right to the voice. Kenerate's Terms give you ownership of what you generate.
Link to this answerRecord or upload a 3–15 second sample of a voice you have permission to use, type the script, and generate: the Qwen3 instant clone reads it in that voice, with no training step. About 10 seconds of clean speech from one speaker works best. To reuse the voice with emotions, speed and pitch, save it once as a MiniMax clone.
Link to this answerIn depth
What works, what to avoid, and how the pieces fit.
Zero-shot cloning models are trained on thousands of voices, so they learn what makes a voice recognisable — timbre, pitch range, accent and speaking rhythm. Given a few seconds of a new voice, the model extracts those characteristics and applies them to new text without any per-voice training [1].
Because the clone is conditioned on the sample, the sample's quality and style carry through: background noise, room echo and mood all come along. That's why a clean, representative 10-second clip matters more than any setting.
Two workflows, for two jobs:
Voice cloning is the same technology behind audio deepfakes — synthetic speech used to impersonate someone [2]. Clone only voices you own or have written permission to use, tell your audience when narration is synthetic where your platform requires it, and never use a clone to mislead.
Free to start. No watermark. Download every clip.
Tried it? Tell us how it went
