Software to trigger a macro based on microphone input (specific sound effect)

Anybody know if there is a good open source program that is capable of triggering a macro (pre-recorded set of mouse/keyboard actions) via microphone detection of a specific sound effect OR at least triggering by tone detection? I know of some closed source programs that can probably do it, but I’d rather see if there are open source options for this.

Personally wouldn’t mind having Alexa/Google action at home w/o web-based solving. I didn’t follow up, but a while back I did a search on DIY speech recognition. The potential solutions out there may work for your application, perhaps.

Check out Mycroft AI and open assistant

In the analog domain a bandpass / notch filter for a tone would work. Also switched cap filter is a possibility. (analog.com)

In the digital domain you’re probably looking at some type of ML - especially if your looking at something more complex than a tone. Might take a look at TinyML. Some links -

1 Like

Trying to get Tensorflow2 installed on Windows because it appears that belwilderingly there isnt an “off the shelf” software available that can do this. Which is almost hard to believe. Cant believe that there isnt some kind of generic “recorded audio recognition” program that can be given an audio file for comparison.

From the last link listed above -

They also have a forum -

There may be a “generic program”, but you will have to train it for your application.

4 Likes

Created an ML model in Edge Impulse but when I try to deploy it (deployment page, green button at the bottom) to my phone or laptop I get the same error. Attached a screenshot

Forgot to add that I have tested audio classification and live classification with the model, so I know the model is working already; but I must be missing some kind of completion step somewhere to enable it to be deployed. Unless this is just a bug

Following… This is interesting! I gots plans for which this could be the exact viable solution! Woot!

Thanks, All! Very much appreciated!

DD

2 Likes

If you dont mind me asking, what project are you working on? Curious to find out how it relates.

Sorry… that’s classified. :wink:

Just kidding. Initially I want to be able to trigger a voice recorder to begin recording based on a verbal command. I don’t trust any of the most omni-prevalent (new word!) voice-controlled apps due to their need to constantly ‘listen’ 24/7/365. Don’t like the constant surveillance. I’ve got IP stuff I’d like to keep under wraps till I decide to publish/produce/fabricate/etc.

Yes… I know I’ve got trust issues. :woozy_face:

DD

2 Likes

I think thats a perfectly reasonable point of view on the principle alone

1 Like

You might find OpenAssistant or Mycroft suitable to your needs. Both of those are capable of operating offline if im not mistaken and are built around privacy concerns

2 Likes

@alepkowski ::
Thank you so much for the recommendations! I’ve put them on my research list. Very much appreciated! You ROCK!

DD

1 Like

“Only the paranoid survive” - Andy Grove

2 Likes

So true. So very, very true.

1 Like

The error mentioned above appears to have been a bug in the edge impulse software. It appears edge impulse engineering team has released a hotfix for this bug. Havent verified it yet

2 Likes

They fixed the bug, so I was able to test some models last night, but every tweak I try I still get a ton of false positives. With so many parameters that I can tweak, i dont know where to first start to try to efficiently eliminate these false positives (falsely identifying the target sound)

1 Like