Voice-Clone Studio

In the Voice-Clone Studio you collect recordings of your voice, check them, process them and train your own voice model from them. You can load several clips (also from a link), tick and untick them individually and apply filters per clip. It's its own window with a sidebar for the tools; you choose the profile at the top under “Profile:”. (The “Profiles” and “Sounds” tabs are in the Voice-Memo Studio.)

“Check clips” measures every clip locally on your computer — free of charge — and gives it a score based on sound, signal-to-noise ratio and reverb. If a clip is too noisy, clipped, too reverberant, contains hardly any speech or is too short, it lands in the red zone, and “Best” never picks it. If you edit a clip after checking, its result is marked as outdated. The next “Check clips” only measures those and clips that haven't been checked yet. Without measurements there's no “Best” — check first. Adding bad clips makes the clone measurably WORSE, not better.

If you accidentally load the same recording twice, the studio recognizes it by its CONTENT (not the file name) and shows the duplicates so you can remove them. Duplicate material skews the training.

These tools run locally and cost nothing: Rumble filter (removes low-frequency rumble), Denoise (smooths out steady background noise), Normalize (raises quiet recordings) and AI denoise (removes noise and surroundings, even while you're speaking). Separating your voice from background music is NOT free: for that there's “Separation · Pro” in the cloud, at 20 credits per started minute. The former local separation was removed on 28 August 2026.

These tools run in the cloud and each cost 20 credits per started minute: “Enhance” (makes dull recordings brilliant), “Separation · Pro” (the best separation of your voice from music and noise), “Remove reverb”, “Remove noise” and “Clearer” (higher resolution). “Separate speakers” costs 1 credit per started minute and trims a conversation down to ONE speaker — ideal for getting a single voice out of a podcast. The price is shown on every tile.

The analysis transcribes ALL recordings of a profile — not just the ticked ones — and evaluates how this voice talks: recurring phrases, sentence structure, contractions, what's noticeably missing. The result comes as a PDF and is meant for prompts and for looking things up; it has no influence on the clone itself. You pay for transcribing the recordings that haven't been transcribed yet (to the second, at least 1 credit per recording) plus the summary at the end. Recordings that are already transcribed are included for free. Both are shown in the analysis window before you start.

With the “Train voice” button you build your own voice model from your recordings. It matches the voice more closely than a single reference recording and afterwards doesn't need one at all — the voice lives in the model. The window guides you through three steps: (1) “Where does the material come from?” – via “Choose folder …”, “Use the checked selection” or “Only the ticked ones”; (2) “Name and size” – the tier follows from the amount of material, you tick three boxes (your rights to the voice and two statements required by law) and start with “Train voice”; (3) “Save speaker” – after that, the voice is available in the “Trained voices” list. The price depends on the amount of material and is shown in the training window before you start. Two things you should know: the training is paid individually in euros, not with credits — neither free nor purchased ones. And it takes a long time — the duration is shown in the training window for each tier. When it's done, the app lets you know with a notice the next time it starts.

How much material is enough is shown in the training window on the amount line. Below a lower limit, no training happens at all; just above it, only with the ‘no quality guarantee’ box ticked; after that, the recommended range begins. Anything above the upper limit isn't included in the training. What matters is quality, not quantity: more material only helps ABOVE the check threshold. Adding clips that failed makes the result measurably worse — that's why the studio warns you when failed clips are ticked.

You can train and keep as many voices as you like on any plan. On the Solo plan, exactly one of them is active at any time: in the ‘Trained voices’ list you switch with ‘activate’, and nothing gets deleted. With Pro and Agency, all of them can be used at the same time. The tier depends only on the amount of material, not on the plan.