After some experience with writing lights FW and playing with some Anduril and other available FWs, especially FW for dual channels lights with one or two switches I realized that it is always very complicated and requires remembering sequences of short and long presses etc.
So I decided to try and develop a voice controlled dual channel light (driver + FW).
My goal is to use it for caving but since dual channel drivers are becoming popular for tint ramping it can be used for any type of light.
After some internet research I decided to choose Speak Recognition Module V3 by Elechouse. it comes with an Arduino library and some examples. Working with the module is plug and play but some YouTube videos make it simple to understand everything.
The module requires 5v so will need some DCDC converter from Li cell and used 2 pins of the MCU so regular 8 pin attiny may not be enough (vcc + gnd + pwm1 + pwm2 + tx + rx + cap + rst + r1 .....).
First you have to train the module using the computer but after the training it is completely stand-alone, no computer or internet connection is needed.
I have trained it to recognize 2 gropes of commands:
a. Main menu:
1. Spot - choose spot channel and wait for input from grope 2
2. Flood - choose flood channel and wait for input grope 2
3. off
4. battery check mode
5. add - next input is in addition to current state (don't turn off current channel)
b. levels menu
1. moon
2. low
3. med
4. high
5. turbo
6. off
After some play with this command I am very happy and sure that with some more research it can "operational" and be used in real lights
Recognition rate is not perfect but maybe some more training is required and also take into consideration that English is not my mother tongue.
in headlamps where the microphone is closer to the moth it might work better.
Here is a short video of my experience:
https://www.youtube.com/embed/LoWbfZWHH30
A lot of work is still needed on the code and electronic circuit. If you have any advice I am happy to hear.
When my code will be more mature I will upload it to my Github.
Well this is fascinating! Nice job!
Can it differentiate between your voice and those of others that may show up in the background?
Might be fun to train it to respond to owl sounds and take a walk in the night woods. 
You can train it to any sound but you will probably have to be close to the Owl.
Not sure about differentiating your voice from others, if they sounds like you or similar to you they will probably can control your lights. As far as I know also Siri etc are not able to response only to your voice and she have a lot more computing power. but you can program a trigger word such as "hey Siri", "OK Google" or "BLF" to prevent accidental control
hey flashy, start fire on other guy’s pocket!
I can’t help but think of an acqaintance who’s favorite exclamation is “f**k Me!!”.
It can be solved. For example only listen/perform command if the switch is pressed. It will eliminate the completely hands free use but solve "funny" friends problem
You’re going to have to get small. Cell phones have had waterproof microphones for many years. IF you could incorporate one of them from an old phone it would help size wise. It would probably not be easy though. You’d have to watch quite a few phone repair videos to try to find a model that you might be able to incorporate.
Waterproof mics are sold on aliexpress for ~1 USD but first I have to make everything run from a battery :)
I believe most/all of the newer 18650 and 21700 batteries with type-c charging built in can be used as a power bank with 5 volts out. And you can still use the normal button terminal for nominal 3.7 volts. Edit, a few of the old micro USB ports on those batteries could do it also.