How on-device voice generation works, and why it stays private
Published · Last updated
Part of the guide: On-Device Voice Guide
On-device voice generation means the work happens in your browser, not on a voice server. When you press Generate in Syllivo, no script and no audio leaves your device, because the voice files were downloaded once and stored on your device. If you are still deciding which voice to use, read choosing a voice by characteristic first.
What happens on first generation
The first time you use a voice set, its files download and are verified, then generation begins in the same flow. You do not need to visit a separate downloads screen first. Later generations reload those cached files instead of downloading them again, so the first chunk can be slower while the files load back into memory.
Why memory is freed after each job
After a job finishes, succeeds, fails, or is stopped, the voice model is unloaded from memory. Your generated audio and your projects stay saved locally; only the in-memory model is freed. That trade-off is deliberate: a normal phone has limited memory, and keeping a large voice model resident would make the rest of the device feel slow.
What "unlimited" honestly means here
Unlimited means there is no artificial character or per-generation quota in the app. It does not mean unlimited memory, unlimited length on every device, or generation that continues after you close the browser. The practical limits are your device's speed and storage, and sections you have already generated are saved so you can resume.
Privacy, in plain terms
Your scripts, your reference recordings, and your generated audio stay in your browser's local storage. There is no voice-generation server receiving them, and there are no ads or purchase tracking in this build. If you clear your browser storage, locally saved projects and audio can be lost, so export finished audio you want to keep.
Frequently asked questions
Does Syllivo send my script to a server?
No. Generation happens in your browser using locally cached voice files. Your script and your audio are not sent to a voice-generation server.
Why is the first chunk of a generation slower?
The voice model loads from cache into memory for the job and is unloaded afterwards, so the first chunk waits for that load. Later chunks in the same job are faster.
What happens to my audio if I clear browser storage?
Locally saved projects and audio can be lost, because nothing is backed up to a Syllivo server. Export finished audio you want to keep.