How to record a sample so the clone sounds like you
One speaker, a quiet room without echo and 30 to 60 seconds of natural speech in the language you plan to voice. Music, phone calls and noise spoil the clone.
Updated November 2, 20262 min read
The clone copies everything it hears in the sample: the timbre, the manner of speaking and also the echo, the hum of a fridge or the music in the background. A better recording gives a closer voice, and the provider can't fix a bad one afterwards.
Where to record
Record in a small, quiet room with soft things around, such as a bedroom with curtains and a sofa. Bare walls, a bathroom or an empty room add echo, and the clone keeps it.
Turn off the TV, music, fan and air conditioner, and close the window.
Hold the phone 15–30 cm from your mouth and keep that distance until the end.
Don't cover the microphone with your hand or a case, and don't record through a scarf or a mask.
How to speak
Speak the way you want the voice to sound in the result. If you need a calm voice-over, read calmly; if you need a lively one, read with energy.
Use your normal pace and volume. Don't whisper, shout or stretch the words.
Read the text on the screen to the end. It contains different sounds and intonations, which the clone needs.
If you make a mistake, tap Record again and start over. A long pause or a cough in the middle ends up in the clone.
Record in the language you will voice
The clone sounds most natural in the language of the sample. A voice recorded in Russian can speak English, but the accent and intonation stay Russian. MiniMax is especially sensitive to this, and the app reminds you about it on the New voice screen. A Qwen 3 voice speaks 10 languages: English, Russian, Chinese, German, French, Spanish, Italian, Japanese, Korean and Portuguese.
Which file to choose
A file works if only one person speaks in it, without music or effects. The limits are 10 seconds to 5 minutes and up to 20 MB, and the app recommends 60 seconds.
| Works well | Gives a poor clone |
|---|---|
| Your voice message recorded in a quiet place | A phone or messenger call: the sound is compressed and the other person's voice gets in |
| A fragment of a podcast or a lecture with one speaker | A video with music or a noisy street in the background |
| A recording made with a dictaphone app | A conversation of two or more people |
| 30–60 seconds of continuous speech | A 10-second recording with pauses |
If the clone still doesn't sound like you
Try the other provider with the same recording. MiniMax and Qwen 3 reproduce the same voice differently, and a Qwen 3 voice costs about 1 token, so it's cheap to compare. A voice can't be retrained: to improve it, create a new one from a better recording and delete the old one in My Voices. Other causes and errors: voice cloning failed or doesn't sound like you.
Contact support
Describe the problem and we will reply by email. Add the file name or job ID if you have one.