Maker intro: building local-first voice synthesis for Mac

Hey PH community ✨

I’m building Sovereign Voice — a Mac app that runs AI text-to-speech entirely on-device using Apple Silicon. No cloud, no API keys, no per-character billing. The model (Qwen3-TTS 0.6B) runs via MLX on the Metal GPU.

Why I’m building this:

I was paying a cloud voice service to synthesize narration for short videos. Paying per character. And one day it hit me: my MacBook has a neural accelerator sitting idle 90% of the time. Why am I sending my scripts to a server, paying for each character, and trusting a third party with my voice data?

The architecture is simple: download the app, it installs a menu bar daemon, the model weights live on your disk (~1.9GB, one-time download). After that, fully offline. Pull the ethernet cable — it still works.

Free tier: 25 generations, 1 voice, basic export.

Pro: $19 one-time, unlimited generations, all voices, voice cloning, studio mode.

Launching on PH soon. Would love feedback from anyone who works with audio, voiceover, or cares about local-first software. Happy to answer questions.