Jacek Dukaj in his book "Perfect Imperfection" (Polish hard sci-fi about alternative physics, mind uploading and enclosed universes) wrote about entities vastly superior to human-level intelligence having an ability to simulate humans to such a degree of precision that they were able to determine what a given human would say or do based on his own internal experience and knowledge in real time. The superintelligent beings didn't scan his brain patterns, they just had multiple simulations of him and chose the one that would best match his actions based only on external cues like position of his hands, body language, speech patterns, etc.
A solution to avoid being read to such a degree that your free will effectively is destroyed was to introduce randomness as a program in your own mind processed in an equivalent of a "cloud VM" - random changes in your body posture, hand movements, weird phrases introduced while you spoke could confuse the internal models of you enough to avoid hyper-predictability.
With a daemon approach and clever integration within applications that handle missing fifo correctly, all you need to do to get rid of the sounds is just stop the service. And it's also configurable, so you can set what actions actually do produce sounds.
This sounds like an idea borrowed from games or other media where doing certain actions results in an audible feedback. This is often done for the benefit of the audience, or to enhance the gaming experience, but I had an idea to implement this in my workspace to somehow influence my brain and make it more enticing to write code and somehow beat procrastination.
I wrote a Python daemon which on startup loads small .wav files into memory to make latency as small as possible, and listens for interrupts as well as reads from a fifo. Other programs can send commands to it to play certain chimes on demand. In ~/.vimrc I added autocommands on certain actions - buffer write, enter/exit Insert mode, enter/exit command line, text change, etc. to send commands to my daemon. Now, when I use vim, I get audible feedback of my actions during writing. Since this is all in a separate daemon done in an UNIXy way, adding support for this in other applications should be easy enough if I want to.
If there was interest, I think that I could clean up the project a little and publish it, including a set of free .wav chimes to use.
YAML, TOML and JSON can be ingested to represent the same data structures internally, it's just a few lines of code to decide which load() function should we use for a particular file. Why not support all three formats in your applications for configuration and just let users decide, which one they want to use? Put a 'config.json' in '/etc/app/conf.d/' and you get the same data, as with 'config.yml' or 'config.toml'. Then users can use whichever format they prefer for the input data.