Create your own custom dictionary

Speakey's dictionary is just a folder of plain text files. You don't have to add words one at a time through the UI — you can author whole vocabulary lists and replacement rules in any text editor and drop them in. Here's exactly how the files are shaped.

Where the files go

• Windows: %APPDATA%\io.speakey.app\dictionary\
• Linux: ~/.local/share/io.speakey.app/dictionary/

Everything Speakey knows lives in that one folder. There are two kinds of file you can author: vocabulary categories and replacement rules.

Vocabulary categories — one .txt per list

Any .txt file in the folder is a vocabulary category. The rules are simple:

• One term per line. A term can be several words — Apex Legends, Baldur's Gate — just keep it on one line.
• Blank lines are ignored. Lines starting with # are comments.
• Save as UTF-8. Case is preserved in the output; matching is case-insensitive, so you only need each term once.
• The filename becomes the category name shown in Speakey. It's lower-case and hyphenated; Speakey title-cases it and turns hyphens into spaces. So proper-names.txt shows up as the "Proper Names" category.

# Teammates and clients
Aoife
Nguyen
Kubernetes
kubectl
Slipwave
Anthropic

Replacement rules — _replacements.csv

Replacement rules are unconditional swaps applied to every transcript. They live in a single file named exactly _replacements.csv. One rule per line, in the form heard,corrected:

• Everything before the first comma is what the engine tends to produce; everything after is what you want written instead.
• Blank lines and # comment lines are ignored.

# heard,corrected
cube cuttle,kubectl
eye tunes,iTunes
get hub,GitHub
my sequel,MySQL

Drop it in — that's it

Save your file into the dictionary folder and Speakey picks it up immediately — it watches the folder and hot-reloads changes, no restart. Edit a file while Speakey is running and the next thing you dictate already reflects it.

Reserved files

A couple of files in that folder are Speakey's, not yours: _stats.json holds the learning data behind word ranking, and learned.txt is the category Speakey fills from corrections you accept. You can read them, but let Speakey manage them — and don't name your own category files with a leading underscore.