VoiceLab is our free, open-source voice studio. It creates, clones and analyzes voices entirely inside your browser — no account, no uploads, no server. It is also the backup narration engine behind SCNviral's videos: if our main voice service is ever unavailable, VoiceLab keeps the channel speaking.
The full VoiceLab app, running live below: designed voices (Kokoro neural speech), script reading, transcription, voice analysis and more. Models download once from Hugging Face and then run on your device — after the first load, VoiceLab even works offline.
First load can take a moment while the speech models download. Your voice recordings never leave this device.
Pocket TTS is a small neural voice-cloning model by Kyutai that runs on ordinary CPUs. Give it a few seconds of someone speaking, type a line, and it reads the line back in that voice — this is the engine that clones the SCNviral history anchor as our ElevenLabs backup. It runs below through sherpa-onnx WebAssembly, still entirely on your device.
English only. The first generation also loads the ~190 MB model — please be patient on slower connections; later generations are quick.
How VoiceLab works, what it costs (nothing), and whose work makes it possible.
Everything on this page runs on your device, in your browser. Reference voices, recordings and generated audio are never uploaded to SCNviral or anyone else — there is no server behind this page at all.
SCNviral narration is produced with ElevenLabs. VoiceLab + Pocket TTS are the studio's independent backup: open models, local machines, no subscriptions — the channel can keep publishing even if a cloud service goes down.
VoiceLab is free software under the GNU GPL-3.0 (see its bundled licenses). The cloning panel uses sherpa-onnx (Apache-2.0). Pocket TTS model weights are by Kyutai, used under the CC-BY-4.0 license.