2
$\begingroup$

I want a microphone to pick up sounds around me (let's say beyond a 3 foot radius), but ignore sounds made at my desk, such as the rustling of paper, clicking a mouse and typing, my hands brushing up on the table, putting a pen down, etc.

How hard would it be for AI to be able to distinguish these sounds from surrounding sounds, such as someone knocking on my door or a random loud sound from further away? How would you implement this? Is it possible that a pre-trained model could accomplish this, and work reliably for most people at their desk? I don't have any experience in AI.

$\endgroup$
  • 1
    $\begingroup$ AFAIK this is quite the task to be starting off with. I don't think you'll be able to easily do this with no experience in AI. $\endgroup$ – Recessive Jan 1 at 0:55
  • $\begingroup$ But do you think it's possible with current AI? Would it be deep learning? Or would the "desk sounds" training data be too random and indistinguishable from other sounds? $\endgroup$ – donut Jan 1 at 1:10
  • $\begingroup$ This is beyond my knowledge, and I imagine it is quite difficult. But, the problem you are describing is called audio segmentation. That should give you some grounds to do your own research on. Here is an article describing some techniques for such: towardsdatascience.com/… $\endgroup$ – Recessive Jan 1 at 2:19
  • $\begingroup$ If you have multiple microphones sound can be localized using signal processing techniques. In practice this might be better than an AI approach. $\endgroup$ – George White Feb 1 at 23:05
0
$\begingroup$

I think if you got the dataset, then a standard 1d convolutional neural network would work to some extent. It's not that there is some property of nearby sounds that it would pick up on. It would just memorize all the sounds that tend to come from your desk. I think the coding part would be pretty standard stuff. But collecting the data will be hard. You have to get a really big labeled dataset of sounds coming from your desk and sounds coming beyond a 3 foot radius. This dataset has to be realistic and representative of the real world. Getting that dataset would be pretty tricky but it is doable if you put multiple microphones in your house in order to triangulate the exact positions of all sounds. It would be like GPS but using sounds waves instead of light waves.

| improve this answer | |
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.