SoundScript: Offline LLM + TTS in a Desktop Studio

It's a desktop TTS studio for Windows and Linux that ships with four engines — two cloud, two local.

---

1. What makes it different?

• Anhad — Local TTS — Kokoro 82M (ONNX) — ~340 MB
• Nuqta — Local LLM — Qwen via llama.cpp — 400 MB – 2.5 GB

Anhad runs entirely on CPU — no GPU required.

---

2. Why offline matters

"I work from a cabin with unreliable internet half the year, and I do client work under NDA where cloud APIs aren't an option."

That's a real constraint. But even with good internet, there's a quieter benefit:

1. No telemetry
2. No cloud storage
3. No uploads

Your scripts, voice settings, and generated audio stay on your machine.

---

3. The workflow

1. Write or paste your script — or let the AI copilot draft it.
2. Pick a voice from 29 curated options.
3. Generate — cloud engines stream via your API key, local engines run locally.
4. Export to MP3 or batch process a folder.

The caching layer is a nice touch: inputs are hashed and checked against disk before hitting any engine. Regenerating the same line is instant and free.

---

4. The takeaway

Offline LLM and TTS aren't just for privacy maximalists or people with bad Wi Fi.

They change the economics of iteration — you stop being precious about regenerating, because it costs nothing.

• No subscription
• Your own API keys
• Local engines free forever

---

5. Links

• GitHub
• SoundScript Docs