epicenter-so
Feature idea: Smart paragraphs and punctuation check
Though it would slow down the generation of text, I often find myself wishing that a block of text gets edited a little bit by AI via punctuation and some paragraphs to make it more readable. I very often have to do this manually afterwards.
There is a cool app on the App Store and Play Store called "Letterly" where you can talk into the microphone, and then you get a transcription and you can also have AI edit it without changing the text. This kind of effect would be amazing.
I don't want the text to be turned into a formal email or stuff like this. Just a quick check on punctuation and empty rows and so on to make a text more readable where it's sensible.
Maybe this could be added as an optional checkbox, with a quick note that it will slow down text generation for the benefit of better-looking text.
Don't mean to speak for @braden-w, but this definitely does not sound like something "easy" to implement, due to variability and ill-definition.
Not saying it is not possible. In fact once Whispering gets the output onto one's clipboard, one could "pipe" it to another AI tool such as GPT-4o for post-processing, with specific instructions. That will consume extra tokens, of course.
Just take a look here at OpenAI's own blurb: https://platform.openai.com/docs/guides/speech-to-text/improving-reliability.
For punctuation and paragraph "rendering", you may also be able to resort to other non-AI techniques.
This would be solvable by the user if output post processing with custom prompts gets added.
This would be solvable by the user if output post processing with custom prompts gets added.
Prompting definitely helps. But to my personal disappointment it does not "solve" it. I have to rely on regex (i.e., "non-AI-based" post-processing) extensively to mitigate it.
Clossing this since custom prompts have been added! For reference, you can configure a user prompt on the transcription level, or even better, configure transformations as demonstrated in the demo or the advanced features section of the README.
