Because SuprFlow's Dictionary does not perform a guaranteed final-text find-and-replace. It sends the correct spelling from a correction entry to WhisperKit as vocabulary guidance before recognition. The speech model may still hear a different word, especially with similar names, background noise, or a language/model mismatch. The “Wrong word” field helps organize the entry in the UI; the app uses the replacement target as guidance, not a deterministic mapping from wrong to right.
That matters for a name in a legal note or client record. If the result must always be checked, treat the entry as a recognition aid and keep human review. The basic Dictionary setup guide shows where to add entries; this guide diagnoses why an entry can exist yet fail to change a particular sentence.
A three-sentence diagnostic
Use a fictional name, not a real client's name. Suppose the app writes “Corwin capitol” when you mean “Corwin Capital.” Add the correction, then dictate the same three sentences with the same microphone and model:
- “Corwin Capital called today.”
- “The Corwin Capital committee meets on Thursday.”
- “Compare Corwin Capital with another firm.”
Record the raw and final text where the history screen exposes them. If only the formatted version is correct, the recognizer and formatter did different work. If none is correct, check the selected speech language, active model, microphone, and how you pronounced the phrase. A setting that says English while you dictate in another language, or an English-only .en model for a non-English sentence, can make a dictionary test misleading. A multilingual test guide covers that case.
Try one correction at a time. The app deduplicates and sorts vocabulary targets into a comma-separated prompt. A large list of similar names can make it hard to tell which entry helped. There is no measured success rate for any one correction, so do not promise perfect spelling after adding a term.
When a snippet is the better tool
If you need a known phrase to expand to exact approved text, use a distinctive Snippet trigger. The app applies snippets after recognition and formatting, just before insertion. That is deterministic text replacement when its trigger matches, but the matching is broad: case-insensitive and tolerant of spaces, hyphens, and some plurals. A common trigger can replace a word you did not intend. Test it in a harmless note and review the final output.
Do not use a snippet to conceal an unresolved recognition problem for names, amounts, or negations. The recognizer must correctly hear the trigger for the snippet to fire, and a broad trigger can create another error. For professional records, confirm the final name against an approved source.
Questions people ask
Is the “Wrong word” field ignored?
It is stored as part of the entry, but the recognition prompt is built from the replacement target. It does not apply a final wrong-to-right substitution. Verify current release behavior if the vendor changes this design.
Will editing a past transcript teach SuprFlow automatically?
SuprFlow 1.0.0 uses explicit Dictionary entries. Editing a pasted sentence does not appear to retrain the speech model. Add the entry and run a new synthetic test.
Can a snippet guarantee a correct client name?
It can insert exact text if its distinctive trigger is recognized and matched. It can also fire unexpectedly. Use human review for important names and keep a controlled list of approved triggers.
Related guides
- Add names and acronyms to the Dictionary
- Use snippets without accidental replacements
- Compare two speech models on the same Mac
- Review raw, cleaned, and formatted history