Local · In-character · No cloud
Shangia
Your stream's voice. Every character.
Shangia turns Twitch redemptions into speech in cloned character voices — and they actually read the message the way the character would. A free, self-hosted alternative to paying by the message.
v1.0.0 · Windows · NVIDIA GPU or CPU · runs entirely on your PC
How it works
Every step runs on your machine. Nothing is uploaded, nothing is metered — a thousand redemptions cost the same as one: nothing.
Everything in the box
Built to give your community the loudest, most unhinged, most in-character soundboard on Twitch.
Reads like the character
Not just their timbre — their cadence. A local GPT-SoVITS model performs each message the way the character actually talks, from their own clips.
Cast a voice from a name
Type glados or spongebob, hit Cast. It pulls their clips off YouTube, learns the voice, and finds their face — in about a minute.
Real emotion, from the text
ALL CAPS yells, AHHHH screams, lmao laughs, 💀 reads as a laugh. The message is performed, not spelled out letter by letter.
Chat-driven chaos
Viewers stack effect tags — [chipmunk] [demon] [echo] [angry] — to mangle the voice. Volume is capped so nobody blows out your mix.
On-stream overlay
A transparent OBS browser source with the speaker's face and a clean caption — styled in a live designer, dropped in with one URL.
100% local. 100% free.
Every voice runs on your PC. No cloud, no API keys, no per-message fees — the whole point of not paying Tangia by the message.
Live in three steps
Run install.bat
One time. It sets up the engines and downloads the voice models (NVIDIA GPU or CPU — it asks).
Open Shangia
A desktop icon opens the panel in your browser. Cast a few characters from clips and preview them.
Wire up the redemptions
Add the overlay to OBS and point StreamElements at it. Viewers redeem, your characters talk.