The captions Signal Lab generates are not generic. They are tuned to how you actually write.
How it learns
You start by picking three to five reference artists whose tone you respect. Their captions inform the system on register, sentence rhythm, and vocabulary choices.
Every time you edit a caption that the chain drafted, the difference gets logged. Over weeks, the voice profile sharpens. The captions stop sounding like a generic music account and start sounding like you.
Where to tune it
/broadcast/voice lets you:
- Add or remove reference artists
- Adjust weights so the artist closest to your tone has more pull
- Test the voice against any prompt to see what it produces
What gets flagged automatically
When you edit or write a caption, the voice check runs in real time. It flags:
- Em-dashes, the most common tell that something was not written by a human
- Voice tells from a watchlist: overused phrasings, formulaic openers, mystical-process language, anthropomorphised gear
@or#inside the caption, because those belong in the Tag and Hashtag fields
Why this matters more than features
Anyone can post. The thing peers and bookers respond to is whether the words sound like a real person running their own thing. Voice DNA is the system making sure that stays true even when you do not have time to draft from scratch.
What you do not have to do
You do not write voice rules manually. You do not maintain a style guide. The system builds one as you use it. Your job is to edit when something does not sound like you, and the next draft moves closer.