She be working late.
Speech-to-text is how you talk to AI now. A thought into a prompt, a meeting into notes, a memo into an email. It starts by listening to you.
In 2020, Stanford ran the same recordings through five big speech systems. Black speakers: 35 percent of words wrong. White speakers: 19 percent. One in five Black recordings came back at least half wrong. (Koenecke et al., PNAS, 2020.)
Twice the errors is twice the fixing, on every sentence, every day, at the door of the one technology everyone says is the future. That is not a glitch. That is a head start, handed to someone else.
Then the quieter part. The polished transcript that says something you never said. You only catch that one against your own memory. That is not a mistake. That is being edited into someone else.
A blaccent is what someone puts on to perform Black speech. Speech software does it in reverse: it takes what you said and hands back its own performance of it.
So the tool that writes down what you actually said wears the word. Reclaimed, and accurate.
BLÁCCENT can fix how a word is spelled. It cannot add a word, drop a word, or move a word.
That is not a setting. It is a check that runs after every pass, word against word, and if anything was added, dropped, or moved, the pass is thrown out. You just watched it run on your own sentence.
It may mishear you. It may not revise you.
If you said it that way, that is how you meant it.
"Finna" stays "finna." Say "because" and it writes "because." No model can know when you want the switch, so BLÁCCENT never decides. When you want a rule, "going to" in work email and "finna" everywhere else, you write it once, turn it on when you want it, and see it every time it fires.
Black English is not broken English. It has rules, and it is the most studied variety of American English there is.
"She be working late" means she usually does. English has no single word for that. The machine does not know the rule, so it hears a mistake.
Google and Stanford asked Black users what happens when their phones get them wrong. They change how they talk to be understood. Every day. BLÁCCENT exists so you can stop.
verified
2026-09-07
Your voice stays on your computer. No account to speak. No server hearing you.
The raw transcript is kept as heard, and the audio stays two days so you can go back to it. What the machine did to your words is never hidden from you.
Anyone who has read a transcript of themselves and met a stranger. Anyone who stopped dictating because fixing the machine took longer than typing. Black speech is where the failure showed. The failure is everyone's.
If you run a company that says "inclusion" and your speech stack gets our people wrong at twice the rate, that is a Malinké Productions conversation.
No beta to join yet. No campaign to back yet. When there is a door, you get it first.
Words by Claude Chat, Malinké Productions.
{{ l.said }}
{{ l.wrote }}
Your browser's speech engine heard you, not BLÁCCENT's, so you are watching one of the systems in the study. The polish pass is a language model told to tidy you up. The fence is the same code. Nothing you said is stored or sent anywhere but that one pass.