When the light is off, the switch is invisible and you have to guess at where it is, or where you need to be walking (hitting your shins on things, or stepping on lego)
Using a non-light based interface means it works just as well when the light is on and off, vs being a broken experience for half of the usecases
Another point, the cognitive load of speaking a command is different than pecking through menus on a phone or finding the right button on the remote/device. It may be easier to just speak out load depending on your state if mind.
I happen to have backed a Kickstarter that will hopefully enable me to do just that ;)
I had to do something similar (minus voice controls) to keep my existing remote's volume controls working with an analog volume device [2].
[0] https://github.com/Pulse-Eight/libcec [1] https://github.com/rhasspy/rhasspy [2] https://rkeene.org/viewer/tmp/cec-volume.cc.htm
For practically anything that involves using code words to tell the computer what to do there are better input methods, like physical switches, or purely automated.
This is slightly off-topic, but in your opinion what are some of the reasons people prefer to use voice input with code words over the other options ?