Custom dictionary matching to incorrect words

[mfosker]

Describe the bug

The post-processing pass over the transcript that applies my imported custom dictionary is substituting dictionary target words into places where they don't belong, corrupting words that were transcribed correctly in the first place.

The live transcription window shows the correct words while I'm speaking, so the speech recognition itself is working. The corruption is introduced after I stop speaking, during post-processing.

My custom dictionary contains the US-to-UK spelling family e.g. mold→mould (molder→moulder, moldering→mouldering, molders→moulders, etc.), and the post-processing seems to be matching these targets against unrelated words rather than only correcting the exact source spellings.

FluidVoice_Dictionary_2026-09-03_14-26 copy.json

Reproduction steps

1. Import a large custom dictionary that includes the mold→mould replacement family (molder→moulder, moldering→mouldering, molders→moulders, moldy→mouldy, etc.).
2. Dictate a sentence containing ordinary words that resemble those targets but are not the source spellings, for example: "I was wondering what was in that folder."
3. Confirm the live transcription window shows the words correctly while speaking.
4. Stop speaking and let post-processing complete.
5. Read the final inserted text.

Expected behavior

Post-processing should only replace exact matches of the dictionary's source words. "folder" and "wondering" are not in the dictionary as source words, so they should pass through unchanged. Final output: "I was wondering what was in that folder."

Actual behavior

Correctly transcribed words are overwritten with unrelated dictionary target words. For the example above, the final text was:

"I was mouldering what was in that moulder."

App Version

1.6.9

macOS Version

MacOS 26.6.2

Architecture

Apple Silicon

Logs or crash report

Screenshots or screen recording

No response

[mfosker]

I'm attempting to work around this by using a massively trimmed down custom dictionary. The one I had (attached) was unnecessarily long and also appears to add a bit of latency to the post processing step.

[altic-dev]

did you add it to custom words or custom dictionary? custom words gets that word and biases it. What i mean is that, similar words might push it harder. Thus similar words might get corrected as well.

so the best way is to add it to custom dictionary - not custom words.

Click "Add manually"

add it here. should work.

[mfosker]

Thanks for the reply.
I had added the words as part of a general US -> UK spelling custom dictionary (since parakeet english spells in american).
That was a very large custom dictionary that I found from another user. It's too many words to add individually. But it's also far more words than I need (there are lots of words that americans spell differently that I won't need for my work, so I removed those). I'll update if I still see the issue now that I have a smaller dictionary.
An alternative would be a UK English version of Parakeet if that exists.
Thanks again for this software, I use it a lot and am really impressed with it.