The module that I am using, "Elechouse Voice Recognition Module V3" can be trained up to 80 voice commands stored to different addresses( EEPROM).
In order to train the module, the Arduino needs the VR3 library, and the example code "vr_sample_code" needs to be uploaded to the board (in my case an Arduino Mega 2560).
Then the serial monitor is used to initialize the tranining sequence and store the sound byte at a different address that can then be recalled. As specified in the datasheet, the Voice Rec module has a maximum of 2 seconds per command.
Demo video of the training process:
One thing that I have noticed, once the module is trained, is that, at least with the supplied microphone, if different commands have similar words, the commands may be interchanged.
For example if I have the following 3 commands:
"Auri Health Update"
"Auri Heart Rate"
"Auri Skin Temperature"
Considering that all these commands start with the word "Auri" it confuses the recognition module, so my current ideas to fix these issues are, in order:
Change the microphone to a professional one, retrain the module and check the results.
Modify the voice commands so that they do not include the same word. As much as I would like to keep the word "Auri" at the beginning, I may have to just leave "Health Update", "Heart Rate", "Skin Temperature".
To improve recognition, analyse how the cadance, pronounciantion and intonation influence the accuracy when waiting for a command.
The following video demonstrates some of the issues that will be encountered throughout development in terms of voice recognition in general and specifically in noisy areas:
Another example with LED activation: