Bring the vocal
Upload audio or paste a URL. The worker normalizes input to clean mono 24 kHz WAV.
Teravoice / Singing Voice Conversion
Change the singer. Keep the performance.
A clean source vocal, a clean reference singer, and a serverless 48GB GPU render path. Teravoice turns the Amphion Vevo2 FM model into a simple studio console.
Live render path
Scale-to-zero
- 01Source vocalWAV, MP3, M4A, FLAC
- 02Reference voiceClean target singer take
- 03Vevo2 FM SVCVast.ai 48GB+ GPU
- 04Output WAVCloudflare R2 public URL
The studio flow
Singing conversion, ready in one pass.
Kits-style simplicity, but honest: the dashboard exposes the source-plus-reference singing conversion path we have actually deployed.
Point at the voice
Use a reference singer take to guide timbre while the source performance stays intact.
Collect the master
Track status, review logs, play the result, and download the rendered WAV.
Built for Indian creators
Regional vocals stay regional.
Bring Hindi, Tamil, Telugu, Bengali, Marathi, Punjabi, Gujarati, Kannada, Malayalam, or English vocals. Teravoice changes the singer timbre while preserving the sung performance; lyric translation is not shown because it is not live yet.
01
Auth
Signup, login, private dashboard
02
Render
Amphion Vevo2 FM-only SVC
03
Storage
Cloudflare R2 input and output files
04
Compute
Vast.ai Serverless GPU, scale-to-zero
For apps and teams
One public dashboard. One token-safe API path.
Hostinger serves the app, Cloudflare R2 stores inputs and outputs, and Vast provisions a 48GB+ GPU only when a render is requested.
POST /api/render/upload
POST /api/render
GET /api/render/{job_id}What is live today.
The product surface stays focused on the production features wired to the deployed serverless GPU stack.
Singing voice conversion
Source vocal plus reference voice through Amphion Vevo2 FM-only SVC.
Dashboard accounts
Signup, login, private dashboard, and user-owned render history.
Developer API
Bearer-token routes for Hostinger or other backend callers.