How to Upload Voice for Cloning
For a voice clone to sound true to life, clean recordings matter. A few minutes of good speech give the best result.
Which recordings work
Best of all is clean, calm speech without noise or music, around 1–3 minutes in total. Videos, voice messages, dictaphone and digitized recordings all work, as long as only the person's voice can be heard. Several short clean clips are better than one noisy recording.
What to avoid
Background music, several voices at once, echo, and heavy noise noticeably worsen the clone. If you have a choice, pick the cleanest clips, even short ones. Less material of good quality beats more of poor quality.
Where to find the voice
Check home videos, birthday greetings, voice messages in messengers, old tapes, and answering machines. The audio track can be extracted from a video. Ask relatives to share any recordings they have.
Uploading
On the training page, add the audio files to the voice section and start the cloning. The result is used only for this AI copy and isn't shared with anyone. The finished clone voices the image's replies in his recognizable voice.
- Clean speech, 1–3 minutes.
- No music, echo, or other voices.
- Look in videos, messengers, on tapes.
- The clone is used only for this copy.
Frequently asked questions
Save the story while it is with you
Create a memorial page in a few minutes — gently, beautifully and with respect for your loved ones. Free forever for the text version.
Create a memorialRead also
How-to
How to Upload Message Histories for the AI
How to export message histories and add them to AI copy training to convey speaking style and facts. On the privacy of living people.
How-to
How to Choose Photos for a 3D Head
Which face photos work for creating a 3D head model in Pomni: angles, quality, and quantity.
How-to
How to Turn an AI Copy On and Off
How to activate an AI copy after training and how to pause or delete it. The family's control over the revival.