VoxShield

Trust-based voice fraud screening demo. VoxShield analyzes a suspicious phone recording in two sequential stages:

1. Synthetic voice detection — is the voice human or AI-generated?
2. Accent estimation — only if the voice is human, estimate which of 4 English accents it matches (American, British, Indian, Australian).

It produces a report with a verdict, confidence score, and risk level. VoxShield is a decision-support tool, not a final judge — boundary-confidence cases are flagged for human review.

This is a hackathon-style demo: the ML models are mocked (deterministic pseudo-random outputs derived from the uploaded file), matching the PRD's two-stage pipeline shape without training real models.

Every analysis is persisted to PostgreSQL so past screenings can be reviewed; the UI shows a "recent analyses" table backed by GET /api/history.

Contents

• backend/ — Express API server (/api/analyze, /api/history) that runs the mock pipeline, stores each report in Postgres, and returns a JSON report
• frontend/ — static HTML/CSS/JS UI (served by a small Express static server) for uploading audio and viewing the report + history
• render.yaml — Render Blueprint that provisions a free Postgres database plus the backend and frontend services

Running locally

1. Postgres

You need a reachable PostgreSQL instance. Easiest local option is Docker:

docker run --name voxshield-pg -e POSTGRES_PASSWORD=voxshield -e POSTGRES_DB=voxshield -p 5432:5432 -d postgres:16

The backend auto-creates its analyses table on startup — no migrations to run by hand.

If DATABASE_URL isn't set, the backend still runs fine (analyses just aren't persisted and /api/history returns an empty list) — handy for quick UI iteration.

2. Install & run

cd backend && npm install
cd ../frontend && npm install

# Terminal 1
cd backend
export DATABASE_URL=postgres://postgres:voxshield@localhost:5432/voxshield
npm start   # http://localhost:4000

# Terminal 2
cd frontend && npm start  # http://localhost:5173

Open http://localhost:5173, upload an audio file, and click "분석 시작" (Analyze) to see the mocked pipeline run, the resulting report, and the updated history table below it.

Deploying (Render)

This repo includes a Render Blueprint (render.yaml) that provisions:

• voxshield-db — a free managed PostgreSQL instance
• voxshield-backend — the API, wired to voxshield-db via DATABASE_URL automatically
• voxshield-frontend — the static-ish UI server

Steps:

1. Push this repo to GitHub (already the case if you're reading this from the repo).
2. In the Render dashboard: New + → Blueprint → select this repo. Render reads render.yaml and provisions all three resources.
3. After the first deploy, open the voxshield-backend service, copy its public URL (e.g. https://voxshield-backend.onrender.com).
4. Open the voxshield-frontend service → Environment → set BACKEND_URL to that URL → save (triggers a redeploy).
5. Visit the voxshield-frontend URL — it now talks to the backend, which talks to Postgres.

Free-tier services spin down when idle, so the first request after inactivity may take ~30s to wake up.