Just now
Automatically Learn from Corrections to Improve Custom Vocabulary
I'd love for Superwhisper to automatically learn from corrections made after dictation, instead of having to manually add every word to the custom vocabulary or text replacements.
Wispr Flow already has a similar feature to some extent, and I think it would be really useful to have something like this in Superwhisper.
The idea is that whenever I correct a word after a transcription, Superwhisper could detect the change and learn from it.
However, I don't think every correction should be saved automatically. An AI model could analyze the corrections and decide whether to save them immediately, wait until the same correction happens multiple times, or simply ignore them.
For example, if Superwhisper keeps writing "Git Hub" instead of "GitHub" and I have to correct it every time, it could eventually recognize the pattern and automatically create a text replacement.
On the other hand, if I correct a word just once and the model isn't confident that it's an actual transcription error, it could wait to see if I make the same correction again before saving it.
It should also be able to distinguish between actual transcription errors and simple sentence rephrasing, to avoid adding unnecessary replacements.
Depending on the type of correction, Superwhisper could either add the word to the custom vocabulary or create a text replacement mapping the incorrectly transcribed word to the correct one.
It would also be nice to have the option to review learned corrections, remove any that aren't useful, and disable the feature entirely in the settings.
I think this would help Superwhisper become more accurate over time by adapting to the way we speak and the terms we regularly use, without constantly having to manage the custom dictionary manually.
Pending