AI Voice Cloning

What it is

Voice cloning builds a voiceover from a recording of you. You give SoeMind a short, clean sample, then type a script, and the result is new speech modelled on your sample, so you don't have to record every take.

The sample decides most of the result. The model follows how you sound in it, including its energy and pace, so a calm sample gives a calm voice. To get a different tone, record a sample in that tone. Results vary with the sample, so always listen before you publish.

How to make one

1. 1

   Open Voice Clone in Studio. Pick a language and read the script shown on screen, or upload a recording you already have.

2. 2

   Keep the sample 15 to 60 seconds long: one speaker, a quiet room, no music or sound effects. Trim it to the cleanest part, listen, and save it.

3. 3

   Type what the voice should say. Write it the way it should sound: short sentences and punctuation shape the rhythm. Spell out years, decimals and symbols, and write names the way they are pronounced.

4. 4

   Leave Delivery style on None to keep the tone of your sample, or pick a preset to steer the delivery. Don't add style notes to the script: the model can treat them as part of the voice.

5. 5

   Read the credit price on the button, then generate. Listen to the result, then download it, add free captions, or use it as the voice of a talking avatar.

About the example above

The Burmese clip is a real recording. All the other languages were cloned from that same voice with SoeMind Voice, with the speaker's consent.

Models, voices and languages

• SoeMind Voice: our own voice engine, which clones from your recording. Text can be in any language, including Burmese.
• Sample: record it in the browser or upload an audio file, then trim it. 15 to 60 seconds of clean speech works best.
• Delivery style: optional presets that steer the read. None is the default and keeps the tone of your sample.
• Where you can use it: Studio on the web, or ChatGPT, Claude and Cursor through the SoeMind connector. The finished audio can go into talking avatars, videos and captions.

What it costs

Credits come from the live price book, the same numbers the Generate button shows. 1 credit = $0.01. Typical examples:

Prices change when provider prices change. You never pay more than the quote; the final charge can be lower. Failed generations are refunded automatically.

Limits to know

• Only clone your own voice, or a voice whose owner has given you documented consent.
• Public figures are not accepted, and neither is using a voice to deceive or impersonate someone. See the AI content policy.
• A clone reflects its sample. Background noise, music or a second speaker in the sample can carry into the result.
• Written direction has limited effect on a cloned voice. Rhythm and pauses follow the script well; strong emotion is harder to direct.
• Generation usually takes a few minutes, and the first run after a quiet period can take longer.
• Each run can sound slightly different, so listen to the whole clip before you publish.

Common problems and fixes

It doesn't sound like me.

Set Delivery style to None and remove any notes from the script. If it still doesn't match, record a new, cleaner sample in a quiet room, with your natural voice and no music.

The voice sounds flat.

The clone copies the energy of the sample. Record a new one at the energy you want back, and use shorter sentences in your script.

A name or word is pronounced wrongly.

Spell it the way it sounds in the script. For example, write SoeMind as So Mind.

Years, numbers or symbols sound wrong.

Write them as they are spoken. For example, write 20 percent instead of 20%, and spell out years in words.

The run failed.

Check the message on the job and try again in a few minutes. If you were charged and got no audio, contact support and we will check it.

Questions