I was paying per character to synthesize my own voice. So I built the opposite.

The moment:

I was editing a 2,000-word narration script for a short video. At $0.18 per 1,000 characters on my cloud TTS provider, that single synthesis cost me $0.36. Not much — until you realize I was doing this 10-15 times per video, 4-5 videos per week. $70-100/month. For my own voice.

The realization:

My MacBook Pro has an M4 chip with a dedicated neural accelerator. It sits idle 90% of the time. I’m paying a cloud service to run a model I could run myself, on hardware I already own, with no per-character fee, no API key, and no one else holding my voice data.

So I built Sovereign Voice.

What it does:

• Runs Qwen3-TTS 0.6B (Apache-2.0) via MLX on the Metal GPU
• Model weights live on your disk (~1.9GB, one-time download)
• After that: fully offline. Pull the ethernet cable. It still works.
• No account, no API key, no subscription, no per-character billing
• Type your script, click synthesize, get audio in 5-9 seconds on an M4

What I learned building it:

1. Apple Silicon is seriously underutilized. The M4 can run a 0.6B model at near-real-time. Most developers don’t think of their Mac as an inference target — they think “cloud or nothing.” That’s changing.

2. One-time pricing feels radical but makes sense. The compute happens on the user’s hardware. There’s no marginal cost per generation. So $19 once, unlimited use. No subscription. People are tired of monthly fees for everything.

3. Privacy is a feature, not a marketing angle. I didn’t set out to build a “privacy-first” product. I built it local-first because the architecture made sense. But the privacy benefit is what people respond to. Voiceover artists who work with confidential material — legal narration, corporate briefings, sensitive interviews — they care a lot about where their scripts go.

4. Distribution is harder than building. The product works. It’s been dogfooded for weeks. But getting people to know it exists is the real challenge. Launching on Product Hunt tonight.

The honest numbers:

• 7 activations so far (mostly me and early testers)
• 0 paying customers
• $0 revenue
• 4 months of nights and weekends

That’s the real version. Not the “I built it and they came” story. The product is ready. The distribution is not. Tonight’s launch is step one.

If you work with voiceover, narration, podcasts, or any audio content — I’d love your feedback. Free to try, $19 one-time for Pro.