A voice studio that runs on your own computer
QuxSound turns text into natural speech and speech into text, with nothing sent anywhere. It is free and open source, with no account and no quota. Music, voice cloning and sound effects are optional downloads.
# Docker, on this machine only docker compose up -d docker compose exec quxsound quxsound models pull docker compose exec quxsound quxsound token # Or from a terminal quxsound models pull standard quxsound serve → http://127.0.0.1:8770, on your own computer.
Your voice work, on your machine.
Nothing leaves your computer
Once the models are downloaded, QuxSound needs no internet. Your text and recordings stay on your machine, and the app looks for updates only when you ask it to.
Free, and yours to read
The code is open source under Apache-2.0. There is no account, no plan and no monthly quota, because there is no server of ours in the path.
Your code already speaks to it
It answers the same API as the voice tools you already use. Point your client at your own machine, swap the key, and use a voice from the local list.
Closed to other web pages
It listens only on this computer and needs a token made for your install. Other sites in your browser are refused, by origin and by host.
Consent comes first in cloning
A cloned voice needs a named owner and their agreement, tied to that exact recording. Withdrawing it deletes the sample, and the next request is refused.
Clear about what it is
Every clip is marked as machine-generated. Terms that come with a model are shown to you before it downloads, and it stays out until you accept them.
Measured, or marked as not.
Measured on an M4 Mac mini with 24 GB of memory. We publish a number here once we have run the thing it describes. Speech and transcription are English only.
| Feature | Speed | What it takes |
|---|---|---|
| Text to speech (Kokoro) | About 3× faster than real time | 0.55 GB of memory, on the CPU alone |
| Transcription (English) | Runs on the CPU | 73 MB download |
| Music | Not tested yet | 6.3 GB download, plus PyTorch |
| Voice cloning | Not tested yet | 3.2 GB download, plus PyTorch |
| Sound effects | Not tested yet | 11.2 GB download, plus PyTorch |
Other downloads. Linux (AppImage), Debian & Ubuntu (.deb), and a Docker image at ghcr.io/quxkit/quxsound. The code is at GitHub. Our code is Apache-2.0, and each model keeps its own licence.
Keep your voice work to yourself.
Free to download. Nothing to sign up for.