AudioLevel

Web-based audio normalization tool. LUFS loudness normalization for music, podcasts, and speech.

Drop files, get broadcast-ready audio. No signup, no BS.

Automatically detects content type (music, speech, podcast), runs multiple processing strategies in parallel, picks the best result. Files delete after 15 minutes.

Run it

git clone https://github.com/miikkis-gh/audiolevel-web.git
cd audiolevel-web
docker compose up -d

Open http://localhost:8081

How it works

1. Upload analyzes your audio (loudness, dynamics, spectral content)
2. Classifies as music/speech/podcast based on silence ratio, spectral flatness, etc.
3. Generates 4-6 processing candidates (conservative → aggressive)
4. Runs them in parallel through FFmpeg
5. Scores each result, picks the winner
6. You download

Conservative processing wins ties. Less is more.

Stack

• FFmpeg + SoX - audio processing and analysis
• ViSQOL - perceptual quality scoring (optional)
• Bun + Hono - backend
• SvelteKit - frontend
• Redis + BullMQ - job queue
• WebGL - that sphere thing

Config

Environment variables in docker-compose.yml:

MAX_FILE_SIZE=104857600      # 100MB
FILE_RETENTION_MINUTES=15
MAX_CONCURRENT_JOBS=4
DISCORD_WEBHOOK_URL=...      # optional, for feedback

Dev

# backend
cd backend && bun install && bun run dev

# frontend
cd frontend && bun install && bun run dev

Tests: cd backend && bun test

License

Apache-2.0

---

made by miikkis