It'd be useful to have the same ability to record recognized noises, the same as we have for speech.
This could apply to the built-in recognizers and/or the parrot integration.
Just as with speech, it opens the possibility for using these recordings for future training.
It could save time by bootstrapping your "pop" dataset if you decide to train a parrot model. (Practically, even if you continue to use Talon's pop recognizer and just want to use Parrot for other noises, you'll probably still need to record pops for Parrot to prevent false positives.)
With parrot, it can be useful for identifying false positives, in the same way that recordings do.
Currently, I'm dealing with some false positives on my "cluck" recognizer, but they only fire when I'm on calls, so it's hard to identify what triggered it. If I could just go through the recordings folder afterwards, it would make identifying the problem much easier.
It'd be useful to have the same ability to record recognized noises, the same as we have for speech.
This could apply to the built-in recognizers and/or the parrot integration.