Your computer can’t run Voicebox. Your browser can.
Lyrevox is a Voicebox-compatible cloud studio. Clone a voice from a short sample and generate speech in a browser tab, on a private GPU in the cloud — no graphics card of your own needed.
Plus applicable tax. 10 GPU-hours included each month. No character limits. Cancel anytime.
How it works
Sign in with your email
We send you a 6-digit code. No password to remember.
Subscribe
Pay $9 a month plus applicable tax; checkout runs on Link, Stripe’s payment service. We set up a private GPU and a storage volume that belong to your account alone.
Open the studio
It opens in your browser and takes about a minute to wake. When you stop working, the GPU switches itself off after 5 minutes.
- 10 GPU-hours every month. Hours count only while your GPU is switched on.
- No character limits.
- Your own GPU and storage, not shared with other customers.
- Your GPU switches off after 5 minutes without activity.
- Extra hours are off by default. When your 10 hours are used, the studio waits for your next billing period unless you switch on extra hours at $1 per GPU-hour plus applicable tax.
- Cancel anytime from the billing page.
As a safety limit, one account can use at most 8 GPU-hours in a single day. Prices are in US dollars and shown before tax; checkout adds any applicable tax and may show the total in your local currency.
Questions
How is this different from Voicebox?
Voicebox is a free, open-source app that runs on your own computer. If your computer runs it well, use Voicebox. Lyrevox runs a Voicebox-compatible studio on a cloud GPU instead, for computers that can’t.
In the browser a few desktop features aren’t available: the global dictation hotkey, the floating dictation window, system-audio capture, choosing an output device, and the MCP server for AI agents. Lyrevox is a separate service built on Voicebox’s open-source code; the Voicebox project doesn’t run or endorse it.
Where do my voices and recordings live?
On a storage volume that belongs to your account alone and is attached only to your own GPU. Other customers’ studios can’t reach it. We don’t use your recordings or voices to train anything. If you delete your account, the volume is permanently deleted after a 14-day grace period.
Why does the studio take a minute to open?
Your GPU is switched off while you’re away, so you aren’t paying for idle time. Opening the studio switches it on and loads the voice models, which takes about a minute. The very first open can take a little longer while models download. After that it stays on while you work.
Which voice engines are included?
Qwen3-TTS (with CustomVoice and VoiceDesign), Chatterbox Multilingual, Chatterbox Turbo and Kokoro. A few engines Voicebox supports are left out because their model licences don’t allow a paid service or aren’t clear.
Is the audio watermarked?
Audio made with Chatterbox or Chatterbox Turbo carries Resemble AI’s Perth watermark, an inaudible mark that identifies it as AI-generated. We leave it switched on. Qwen3-TTS and Kokoro output isn’t watermarked.
What happens when my 10 hours run out?
The studio won’t wake again until your plan renews, so your bill stays at $9 (plus applicable tax) unless you choose otherwise. If you’d rather keep going, switch on extra hours in your dashboard: they cost $1 per GPU-hour plus applicable tax and appear on your next invoice. You can switch them off again at any time.
Whose voices can I clone?
Your own, or voices you have clear permission to use. Impersonation, fraud and harassment are not allowed; the acceptable use policy has the details.