🎤
Clone from one short clip
Six to twenty seconds of clean speech is all it needs. No training run, no dataset — pick the file and it is ready.
Give it 6–20 seconds of clean speech and a script. It reads the whole script back in that same voice, pausing and laughing where you mark it, and saves a WAV and an MP3. Everything runs on the laptop — nothing is uploaded.
One voice sample, a whole script in that voice
Six to twenty seconds of clean speech is all it needs. No training run, no dataset — pick the file and it is ready.
Write (pause), (pause 2), (laugh), (giggle), (sigh), (gasp), (hmm), (cry) or (cough) in the text and the voice does exactly that, right there.
English, Hindi/Urdu, Arabic, Spanish, French, German, Italian, Portuguese, Polish, Turkish, Russian, Dutch, Czech, Chinese, Japanese and Korean.
A long script is read in pieces and saved after every one. Press STOP and the part already recorded is still a finished file.
Once the voice model is downloaded, nothing leaves the machine — no account, no API key, no upload of anyone's voice.
Every run writes a timestamped WAV, plus an MP3 when ffmpeg is around. One button opens the output folder.
Four steps, once per laptop.
Open the download link. It opens on Google Drive — press Download, then "Download anyway" if Google asks.
Put the .exe in a folder of its own, somewhere permanent, and double-click it. It shows a Device ID.
Send that code on WhatsApp. Your key comes back within minutes.
Paste it in, press Activate. The first GENERATE then downloads the voice model (~2 GB), once.
When you start the tool it shows a Device ID like this:
PC-4KTM-9QXAThat code comes from your motherboard, and your key is built for it. A key made for one computer will not run on another, so it cannot be passed around. Reinstalling Windows does not change it — your key keeps working.
✓ Lost your key? Send the same Device ID on WhatsApp and get it back.
Two things: one audio file of the voice (6–20 seconds, clean, no music or background noise) and the script you want read. Pick the language your script is written in and press Generate.
Six to twenty seconds works best. Longer is not better — what matters is that it is clean, one person speaking, at a normal pace, with no music or echo behind it.
The first time you press Generate, the voice model is downloaded — about 2 GB. It is saved in a models folder next to the program, so this only happens once per laptop. Every run after that starts straight away.
Only for the download and that first model fetch. After that it runs completely offline — nothing is uploaded, and no account or API key is involved.
No. It uses the GPU if the laptop has one, otherwise it runs on the processor. On a CPU it is slower, so a long script takes a while — the status line shows how many parts are done and roughly how long is left.
The program copies fine, but the key will not work there. Each key is tied to one motherboard. Run it on the other laptop, send me the Device ID it shows, and I will send that laptop its own key.
An output folder next to the program, named by date and time, for example voice_20261017_143500.wav. There is an "Open output folder" button in the window.
Yes. The script is read in pieces and the file is saved after each one, so pressing STOP leaves you a finished file containing everything recorded up to that point.
No. Google Drive does not virus-scan files this large, so it shows that notice for every big download, whoever uploaded it. Press "Download anyway" and carry on.
Yes. SmartScreen warns about any .exe it has not seen many times before. Click "More info" then "Run anyway". The tool runs only on that machine and sends nothing anywhere.
Yes. The Device ID comes from the motherboard, so reinstalling Windows does not change it and the key keeps working.
Download it, run it once, and send me the Device ID it shows. Your key comes straight back.