Describe the voice you want. Hear it in about twelve seconds. Buy the one you like for $50 and download the file.
Nobody recorded it, so nobody can revoke it.
Clone a voice on any of the large platforms and what you are given is a voice ID on their server. It works while you pay. The number of voices you may keep is capped by your plan. There is no export: BreezeBlue’s own documentation says it stores a trimmed excerpt of what you uploaded rather than your file, and download exists only for previews. Resemble bills each voice by the month. Voice.ai counts them against your tier.
In July 2025 Meta hired Play.ht’s team. The API went dark two weeks later, in the middle of customers’ integrations. The service closed on 31 December 2025 and all user data was deleted. There was no export tool. Every voice a Play.ht customer had built is gone.
A Vopus voice is a WAV on your disk. Our servers are not in the loop after the download.
| Service | What you hold | When you stop paying |
|---|---|---|
| ElevenLabs | A voice ID in your account, in their library. | The account is what has it. |
| Resemble | A clone rented at $2 to $5 a month, per voice. | The voice goes away. |
| Voice.ai | A slot; 5 to 2,200 per tier. | Slots go with the tier. |
| BreezeBlue | A voice ID; they keep a trimmed excerpt, not your file. 5 to 300 slots by tier. | Not stated in their documentation. |
| Play.ht | Nothing. Closed 31 December 2025. | All user data deleted. No export. |
| Vopus | A 24 kHz WAV and its papers, on your own disk. | Nothing changes. |
Plans and terms as published by each service, checked August and September 2026. They change; read the current ones.
One voice, paid once. Four things come with it.
A 24 kHz mono WAV of your voice reading the line you auditioned it on. The build you heard is the file you get; nothing is re-rendered after you approve it.
A one-page record that travels with the file: how it was made, that no recording of any person was used at any stage, what it measures, and the file’s SHA-256 digest.
A serial number written into the audio below the threshold of hearing. It reads back after normalising, resampling, trimming and mp3 encoding down to 64 kbit/s. Anyone can check a file at vopus.io/verify. It carries no name, does not call home and limits nothing.
Each purchase adds ten builds to your account, so the next voice starts from a full session.
A voice file needs an engine to speak new lines, and engines are getting cheap. Speechify charges $6 to $10 per million characters. ElevenLabs charges $50 to $100. Chatterbox is free, open source, and runs on your own machine.
When speech costs that little, the part worth paying for is which voice it is and whether you may use it. A Vopus voice settles both. It is a file you hold, and there is no performer behind it whose permission you would need. Load it into whichever engine you like as the reference voice, and change engines whenever a better or cheaper one appears. The voice comes with you.
Building is free. Paying is the last step, and only for the voice you keep.
Age, sex, size, warmth, pace, in plain words. Six tuning controls fill themselves from what you wrote and show their numbers.
About twelve seconds later the voice reads your audition line. The line is yours to change.
Move a control, build again. Up to ten versions sit side by side on one page, each with the settings that made it.
$50 through Stripe. The file and its papers unlock for download, and a receipt arrives by email.
Each of these was built from the description under it and nothing else. They all read the same line, so what changes between them is the voice.
“Nobody recorded this voice. It was built from a description, and it belongs to whoever bought the file.”
These six were built through the same controls you get, with nothing added afterwards. Your own description is the seventh: build it.
To make your voice say something new, load the WAV as the reference voice in a cloning engine. Each one asks you to confirm you hold the rights to the voice you upload. You do, and the papers say why: no person was recorded. Vopus runs none of these engines and is affiliated with none of them.
A model you have downloaded cannot have a new condition added to it later. This is the durable route.
ElevenLabs Professional Voice Cloning and Speechify verify consent by having the speaker read a phrase aloud. Nobody can speak in a synthesised voice, so those two will not take any synthetic reference, ours included.
The papers state the facts a buyer, a client or a platform will ask about: the voice was synthesised from a description, no human recording, donor corpus or reference audio was used at any stage, and here is what the file measures. The SHA-256 digest ties the papers to one exact file.
The mark covers what the digest cannot. Once a file has been trimmed, normalised or converted to mp3 the digest no longer matches anything, and the mark still reads. Drop any file on vopus.io/verify and it answers with the serial number and the date it was issued, or says the file carries no mark.
Ten builds on a new account, no card. Hear your own descriptions, tune them, compare up to ten side by side.
The WAV, the Voice Authorship Papers, the Vopus mark, and ten more builds. Paid through Stripe.
Nothing owed to us. The engine you choose to speak new lines may charge its own rates, or cost nothing on your own machine.
No. The voice is synthesised from your description. No recording of any person is used at any stage, no donor and no reference audio, and the papers record that. It resembles no one by design.
A 24 kHz mono WAV of the voice reading your audition line, usually five to ten seconds, and the papers as a page you can print or save. That WAV is the reference file every cloning engine asks for.
There is no performer, so there is no performer’s permission to obtain. What you may do with the speech an engine produces from it is set by that engine’s terms, which is why the free, open-source ones are listed first.
You keep the file. It never contacts our servers, the mark is inside the audio, and the papers carry the digest. Nothing you own depends on us being online.
Cloning engines want a few clean seconds of one speaker, and a Vopus file is exactly that. Length is for the engine to add. Your audition line can be any sentence you like.