How to organize a continuous listening to the microphone and a reaction to a certain word?

Question

linux

There is a piece of iron in the form of Raspberry Pi 3. It costs raspbian and there is a task to constantly listen to the microphone and when they say a phrase, react (run a script, for example). What is the best way to organize it? If there is no internet connection?
Okay, I'll clarify. There is Internet, but the problem is that I can’t feed it a 24-hour recording (the speed is so-so, traffic is limited). Therefore, I want to teach him to recognize my voice with one single command (address, something like Ok Google), which will already run a python script (mine), which will record the rest of the text and recognize it through Google or Yandex services

Answer 1

Without the Internet, only if you have a budget of a couple of million greenbacks for a server farm and pay for the work of a herd of programmers to write voice recognition.
If in the conditions of the task there is a clause to react only to specifically given words that are pre-set for a pre-known voice and intonation, then there is a solution. For example, CMU Sphinx (sphinx), but frankly speaking, it works a little strange with Russian and generally starts to work very poorly when the number of commands goes over a couple of dozen (at least it was like that when I watched it for a long time). There didn’t seem to be a port for armies, but most likely there are good people.
There is also the easiest option, to use Yandex\Google\... services. Of the advantages, a convenient api, huge accuracy (against the background of any offline solutions). Of the minuses, you need an Internet and free restrictions on the number of requests.
You can also use an android for this case (I don’t remember exactly, but it seems that recognition services go without restrictions but require an Internet).