I’m the new guy, really new, and more of a hobbyist when it comes to the technical aspect but I have an idea that I’ve been kicking around for a while now, so this idea led me to find Mycroft.
Has anyone successfully used sound or sound frequency as wake “words” apposed to spoken words? Can wake words also trigger a command or a list of commands? Is this even possible?
For example: When my doorbell rings I want to execute a commands list based on the sound or frequency of the doorbell.
Interesting, Just hypothesizing here but you may be able to utilize snowboy as hot word detection and train the model on your doorbell. In mycroft there is likely a method to determine which hotword engine was triggered thereby triggering your “commands” only when snowboy is triggered thus still having “hey mycroft” operate as expected. This will definitely require digging into the code a bit.
Hi @dmorr, great question.
Some Wake Word listeners, such as PocketSphinx, are word-based. Our default Wake Word engine, Precise, runs using a neural network. Theoretically it could be trained on a sound sequence rather than words, however currently our training of Precise is limited to the Hey Mycroft Wake Word.
Alternately, until a completely software option is found, a more hardware-based option could be possible. Add a relay to the doorbell circuit (making sure it’s for the correct voltage and correct AC/DC), then an opto-isolator between that and an input pin on a microcontroller/pi(zero?)/whatever, and run a script on that that when the pin is triggered, it bypasses the STT entirely and simply sends a message to Mycroft such as Hey Mycroft doorbell , at which point it’s up to a Mycroft skill to respond as necessary (which it eventually would be anyways with any approach, of course).
Also, if you’re uncomfortable with wiring to a doorbell circuit, this person seems to have set up a Pi to flash his smart lights when it hears the doorbell, and that might be adjustable to send commands to Mycroft when triggered, but it doesn’t sound like it discriminates between a doorbell and any other loud noise near the mic very well, so YMMV with that approach.