LuminAudio

Demo: https://luminaudio.vercel.app

An open-source web frontend for generating and downloading high-quality, long-form text-to-speech audio. Built for audiobooks, voice-overs, narrations, and anything else that needs great-sounding speech at scale.

Powered by Resemble.ai Chatterbox, running entirely in your browser/local environment. Your API key never leaves your own server.

---

Features

• Long-form TTS — up to 50,000 characters per generation; long texts are automatically split at natural sentence/paragraph boundaries and stitched back together
• 7 built-in voices (3 male, 4 female) + support for any custom Resemble.ai voice UUID
• WAV download — lossless output at your chosen sample rate and bit depth
• MP3 export — client-side 128 kbps conversion, no server round-trip
• Configurable audio quality — sample rate (8 – 48 kHz), PCM precision (16/24/32-bit or MULAW)
• Expressive controls — speaking pace, exaggeration, and temperature sliders
• Auto model fallback — silently retries with chatterbox if chatterbox-turbo is unavailable for a voice
• Fully client-side MP3 encoding (no server processing of the final audio)
• Responsive, accessible UI — keyboard navigable, ARIA-labelled, reduced-motion safe

---

Tech Stack

Layer	Technology
Framework	Next.js 16 (App Router)
UI	React 19 + Tailwind CSS 4
MP3 encoding	@breezystack/lamejs (client-side)
TTS API	Resemble.ai Chatterbox
Language	TypeScript 5 (strict)

Key Dependencies

next                   ^16.1.6   — React framework
react / react-dom      ^19       — UI rendering
@breezystack/lamejs    ^1.2.1    — Client-side MP3 encoding (lamejs fork, Turbopack-compatible)
tailwindcss            ^4        — Utility-first CSS

---

Quick Start

Requirements: Node.js 18+, a Resemble.ai account with an API key.

git clone <repo-url>
cd luminaudio
npm install
npm run dev

Open http://localhost:3000, paste your Resemble.ai API key, and start generating.

Production build

npm run build
npm start

---

Getting a Resemble.ai API Key

1. Sign up at app.resemble.ai
2. Go to Settings → API and copy your API token
3. Paste it into the API Key field in LuminAudio

The key is sent on each request from your browser to the Next.js API route, which forwards it to Resemble.ai. It is never stored, logged, or persisted anywhere.

---

Voices

Name	Gender	UUID
Lothar	Male	78671217
Orion	Male	aa8053cc
Ash	Male	ee322483
Hem	Female	b6edbe5f
Lucy	Female	fb2d2858
Abigail	Female	91b49260
Cutesy	Female	5ec517ba

You can also enter any Custom UUID from your own Resemble.ai voice library.

---

Audio Settings Reference

Setting	Range	Default	Description
Sample Rate	8000 – 48000 Hz	48000	Output audio sample rate. 48 kHz is CD-quality and the recommended default.
Precision	PCM_16/24/32, MULAW	PCM_32	Bit depth of the WAV output. PCM_32 is highest quality; PCM_16 is broadly compatible.
Speaking Pace	0.5 – 2.0	1.0	Speed multiplier. 1.0 = natural pace.
Exaggeration	0.0 – 2.0	0.65	Controls emotional expressiveness. Higher = more dramatic delivery.
Temperature	0.0 – 2.0	1.3	Sampling randomness. Lower = more consistent; higher = more varied.

---

Security Notes

For local use (default)

No extra configuration needed. The app is designed for personal, local use. Your API key travels over localhost only.

For remote / shared deployments

Warning: LuminAudio has no built-in authentication. Anyone with access to the URL can use your Resemble.ai API key to generate audio, incurring costs on your account.

If you're deploying this for others to use, consider:

1. Serve over HTTPS — the API key is transmitted in request bodies. Always use TLS in production. The Strict-Transport-Security header is already set.
2. Add authentication — protect the app behind a login (e.g. HTTP Basic Auth via a reverse proxy like nginx/Caddy, or NextAuth.js).
3. Use a server-side environment variable instead of client-provided keys:
   • Set RESEMBLE_API_KEY=your_key in your environment
   • Modify src/app/api/tts/route.ts to read process.env.RESEMBLE_API_KEY instead of accepting apiKey from the request body
   • Remove the API Key input from the UI
4. Rate-limit the /api/tts route — e.g. with a reverse proxy or an in-process rate limiter to prevent abuse.

What's already hardened

• Server-side input validation: text length (≤ 50,000 chars), API key length, voice UUID length, numeric param clamping, model/precision/sample-rate whitelisting
• Security headers on all routes: X-Content-Type-Options, X-Frame-Options, Strict-Transport-Security, Referrer-Policy, Permissions-Policy
• Upstream error messages are truncated before being forwarded to the client
• No cookies, no sessions, no database — stateless by design

---

Project Structure

src/
├── app/
│   ├── api/tts/route.ts   # Server-side TTS proxy → Resemble.ai
│   ├── page.tsx           # Main UI
│   ├── layout.tsx         # Root layout
│   └── globals.css        # Global styles + Tailwind
└── lib/
    ├── mp3.ts             # Client-side WAV → MP3 conversion
    ├── wav.ts             # WAV header parsing and concatenation
    └── chunker.ts         # Smart text splitting for long inputs

---

License

MIT — free to use, modify, and self-host.