Why Doctor Djeli doesn't take voice notes yet.

Customer outcome

We wanted you to be able to talk to Doctor Djeli instead of typing. So we tested a speech-to-text service with a ten-second recording that said her name three times. It wrote 'Dr. Jelly' three times, with no flag and no question mark. It was fast and it was free, and it changed her name. The evaluator still asks you to type.

Who it serves

Anyone whose voice has been corrected by a machine. That includes most of us. Speech recognition gets Black speakers wrong nearly twice as often as white speakers, and a tool that turns 'finna' into 'going to' did not fix anything. It translated you without asking. BLACCENT, a speech product from our studio, Malinké Productions, is being built on the opposite promise: it may mishear you, but it will not revise you.

Cost or trade-off

Typing is slower, and some people would rather talk. We chose slower over rewritten.

Still not right

The evaluator still asks you to type. When voice comes, you will see the transcript and fix it before anything is sent.

Denominator

1 test recording. 3 of 3 mentions of her name changed.

Commit / deployment

Typing-only evaluator read back in production commit 6e426cfee2bc2b0fc6a694c6d41152cfeda0375a, deployment dpl_HEiJS8aKG8jhYsoPDYKekBieMDPh (READY); approved release candidate e6d05b8813437285bfd2dda93ce3d339779d3a00.

Responsive witness

The service's response was kept exactly as returned, 26 August 2026, with no edits.

Canonical readback

https://rootandreign.co/evaluate returned HTTP 200 on 24 Sept 2026; the served page has no microphone control.

Registry readback

The voice build is held at provider readiness.

Held work

• Codex — A transcription layer that keeps 'Doctor Djeli' intact, and shows you the text to edit before it is sent.