Apppublic

RishiKar210/voice-normalization

sourceHugging Facemitupdated 3mo agoView on Hugging Face

0likes

App README

Voice Normalization Backend

FastAPI backend for the voice-normalization tool. Handles uploads, speaker diarization (via ElevenLabs Scribe), voice cloning, S2S conversion, and per-speaker track export.

The Vite frontend lives separately on Vercel and proxies /api/* here.

Required secrets

Set these in this Space's Settings → Variables and secrets:

Name	Required	What it does
`ELEVENLABS_API_KEY`	Yes	All speech work (S2S, IVC, TTS, Scribe, Voice Isolator)
`ALLOWED_ORIGINS`	Recommended	Comma-separated list of allowed frontend origins, e.g. `https://your-app.vercel.app`. Defaults to localhost for dev.

Endpoints

Health check: GET /api/healthz → {"status":"ok"}

Full API docs: GET /docs (FastAPI auto-generated).

Storage note

Free-tier Spaces have ephemeral storage — uploaded clips and run outputs disappear when the Space restarts (auto-sleep, redeploy, etc.). The voice bank ships in the image and is always present.

Add HuggingFace Persistent Storage (~$5/mo) if you need clips to survive across Space restarts.