Vocabulary, replacements and smart match
Three ways to stop Blurt mangling the words that matter to you: bias the model toward them, rewrite them after the fact, or let it catch near misses.
Custom vocabulary
Add names, product terms and acronyms in Dictionary. They are weighted so the engine favours them over similar-sounding ordinary words. Keep the list to terms you actually say — a hundred rare words makes recognition worse, not better.
Word replacements
A heard-form and a replacement, applied to every transcript after recognition. Use these for the ones the model keeps getting wrong in the same way.
kubernetes → K8s see quel → SQL
Smart match
Fixes near misses of names already in your vocabulary, without you writing a rule for every misheard variant.
Corrections from history
Edit a transcript in History and Blurt keeps the correction, so the same fix does not have to be made twice.