Let This Crying Detecting Classifier Offer Some Much Needed Reprieve

Baby monitors are cool, but [Ish Ot Jr.] wanted his to only transmit sounds that required immediate attention and filter any non-emergency background noise. Posed with this problem, he made a baby monitor that would only send alerts when his baby was crying.

For his project, [Ish] used an Arduino Nano 33 BLE Sense due to its built-in microphone, sizeable RAM for storing large chunks of data, and it’s BLE capabilities for later connecting with an app. He began his project by collecting background noise using Edge Impulse Studio’s data acquisition functionality. [Ish] really emphasized that Edge Impulse was really doing all the work for him. He really just needed to collect some test data and that was mostly it on his part. The work needed to run and test the Neural Network was taken care of by Edge Impulse. Sounds handy, if you don’t mind offloading your data to the cloud.

[Ish] ended up with an 86.3% accurate classifier which he thought was good enough for a first pass at things. To make his prototype a bit more “finished”, he added some status LEDs, providing some immediate visual feedback of his classifier and to notify the caregiver. Eventually, he wants to add some BLE support and push notifications, alerting him whenever his baby needs attention.

We’ve seen a couple of baby monitor projects on Hackaday over the years. [Ish’s] project will most certainly be a nice addition to the list.

45 thoughts on “Let This Crying Detecting Classifier Offer Some Much Needed Reprieve

  1. This is a terrible idea, and I don’t think any parent worth their salt would be happy only being alerted to a baby’s distress with that accuracy. This post is bad and you should feel bad for mindlessly amplifying bad noise.

      1. 86.3% leaves a rather large margin for something to go wrong. Also… I don’t know how they take care of their kid, but what if the sound it’s chosing to reject is the baby spitting up while laying on its back? Not a crying noise, don’t need to send an alert. That’s a dead kid. The only reliable monitor for babies is one that sends everything so the parent with a functional brain and sort through the data and decide what needs dealing with and what doesn’t. Again I can’t say how well they take care of their kid, but I certainly wouldn’t trust a human life to 86.3% accuracy.

        1. Who is the “they” you refer to here? This project is a simple example of a potential application of audio classification – there is no “trusting a human life” happening here.

          > The only reliable monitor for babies is one that sends everything

          This doesn’t exist. All baby monitors use at least a volume threshold to decide whether to transmit audio – what you describe is “always on” which is not how these devices work.

          > so the parent with a functional brain

          This also doesn’t exist. Babies take months to develop regular sleep cycles, causing extreme sleep deprivation to parents, impairing brain function.

          1. Cool, so your not going to trust a kid to your project, awesome. Maybe ask the author of this article to make that clearer, because it reads like you intend to use it in real life. And yeah sure we could go to your project page and probably find that out, but that doesn’t change that article doesn’t make that clear.

            “They” was used because I wasn’t sure what gender pronoun you go by.

            Also, when I said functional brain I meant exactly that. We all get some loss of sleep with babies but I have never found myself unable to make clear and we’ll reasoned decisions because of it. Maybe sleep on those weekend nights, then you won’t feel so impaired. Being awake and functional makes parenting alot easier, so maybe stop wasting those weekends.

            I don’t know why you’re baby monitor filters anything. Mine certainly doesn’t, and it’s one of those fancy ones with video feed. The only losses are from the digital compression, and even then it’s just white noise that gets lost

        2. 100% agree, came to say the same. This is exactly the, in my personal opinion, horrible use cases of ML. Sure, it’s nifty 85%, of the time (or even if it was 98%), but what if the noise is choking, etc. And re: the below, yes, this reads as full intent to be used in real life (unless I skimmed too fast), not as a ‘for fun’ project. If you want a for fun project, how about something not involving neglecting an infant/toddler as the result of poor ML classification. Same crap applies in many areas – loan processing – sure, it seems great – lower loan processor labor costs, unless you happen to be the poor bastard that ML denied your loan application because you fit some oddball pattern ML picked up on, quite possibly due to inept training letting a model pick up on it in the first place, that had nothing in reality due to your ability or likeliness to hold up your end of the deal.
          – Sometimes humans make bad calls, but then at least it’s human judgement which is expected to be fallible occasionally – machines do not get this benefit of the doubt. Good luck explaining to a judge your infant/child died because you were relying on a ML model to alert you if they made noise worth checking on, and it didn’t classify the said event as critical.

          1. – And I get this may be ‘in fun’ – but the issue is still others may follow, and especially as revisions progress and seeming error rates decline, it is easy to start slowly trying and relying on what was ‘for fun’, even if it is just others who do the same. Then maybe a jump to an investor with more money than conscience or brains to try to market it, especially if detection stats (even if only for crying rather than the real importance of ‘critical situations’) are made to look nice.
            – I guess I have a pet peeve for ML models which have possible negative outcomes for human lives, which could have easily been prevented simply by not trying to ML model it in the first place and just let a human process the info – more jobs, and likely less errors, especially on less common scenarios that may not train well. – More to the loan processing side, but ‘the computer said so’ is also much less questionable/appeal-able than a human-to-human discussion or decision – and similar applies elsewhere, where the computer is assumed/expected to be infallible.
            – Like with some of the facial detection / lookup softwares, ‘be careful what you build and release into the wild’ – and sorry to take this out on your project – glad you’re having fun learning ML ;-)

          2. No, it’s far worse than ML rejecting a loan.

            If ML rejects a loan, they can appeal, and a human can re-check.

            If this misclassifies coughing, you potentially have a dead baby. Parents – mothers in particular – have an amazing ability to filter their baby’s noises, even whilst asleep.

            This might be fine to use with an older kid, but not a baby.

    1. Thanks for the feedback! Perhaps it’s not clear, but this is just a proof of concept – not a commercial product. The problem with conventional baby monitors is that they are activated by any and all noises, meaning that they go off unnecessarily all day and all night, which can be incredibly frustrating for sleep-deprived parents, woken at 3am by a fan or other unimportant sound. The point of the project was to experiment with noise classification in an application where it might really make a difference with vs. traditional solutions, and the “dumb” baby monitor seemed like a great target to make “smart” with machine learning. Kudos for the Zoidberg reference though! :)

      1. How did you have time to do this? My daughter is now a year old. For the past year, I have either had no personal time or I have been so freaking exhausted that I wouldn’t have been able to do this.

      1. Thanks for your comment! I’m not sure if you had a chance to read the project, but it explained that this is the first part in a series, throughout which the issues you mention will be addressed. :)

      1. It should be possible to tune the classification to prefer false positives over false negatives. This way less real problems are ignored and still some events not needing reaction can be filtered out.

  2. Just have the baby sleep in or next to your bed – they tend to feel much safer then and won´t cry that much.
    and for during the day there are “backpacks” you can mount them in front of your tummy/back. That way they get much movement and emotional warmth and won´t bother you much while you still can mind your business.
    they are well adopting to oise, as long as they feel safe they will sleep fine eve in noisy environments.

    1. Seriously? No. Babies should NOT sleep in bed with full sized adults. The risk of suffocation or injury is real.

      “U.S. medical groups warn parents not to place their infants to sleep in adult beds due to serious safety risks. Bed-sharing puts babies at risk of suffocation, strangulation, and SIDS. Studies have found that bed-sharing is the most common cause of deaths in babies, especially those 3 months and younger.”

      1. Works fine in most countries. In the U.K. it’s safe unless parents are drunk/drugged or seriously obese.

        So yeah, may not be a great plan in the US.

        You can however pop a cot next to the bed – some are designed with a drop side At bed height so you can easily transfer the baby back after feeding.

    2. Seriously, don’t sleep in the same bed with an infant. Many kid have died this way. Wait till their older! Or if you’re wise, never make it a normal thing. Other wise they think they can all the time. Our kiddo is almost two and he’s never been allowed to routinely sleep in our bed. Because of that he sleeps in his bed 95% if the time without any issues and even puts himself in his bed when we say it’s bed time. He doesn’t like that it’s bed time, but he doesn’t fight about being in his bed. He understands that he’s supposed to sleep in his bed.

      1. Hmm see your point there. Ours is 6 now and crawls over in 23% of the nights (rate is slowly dropping) 2h before it´s wake up time. If she decides to sleep in her bed that is (80% of the time). got a bigger bed so no problem. Intention was to have her cutting the cord when she feels like it so she always feels as save as she needs. Never worried about suffocation/strangulation as we´re not moving much and the bed is big enough. and some things nature will sort out, how tragic it may be, there calculatable risks in everything we hack together.. even though wearing earplugs i always woke up when she had an issue, feels like our neuronal networks are connected in some fascinating hive way.
        or we were just stupid and had luck:) sleeping in her bed she would always wake up in the night, sleeping near us, she never woke up. as soon as she could crawl she would just find her way to the milk and drink it (before she simply slept next to the milk glands) like other mammals offspring does. great for the mommy so she could feed while sleeping:)
        your milage may vary with a child that does not sleep the whole night.
        but had i sleep her in a different bed or room, i would´ve totally gone for this really nice project here!

  3. The negative comments (as well as “just be this kind of parent and you won’t need it”) surprises me. Great project, very interesting problem and solution! I can imagine it being useful, since baby monitors are in fact useful things, and I think all parents understand that electronics can fail and shouldn’t be relied on in life or death situations. Why is a DIY solution less reliable than a unknown-brand, unknown-quality, lowest bidder type product, sleekly packaged with a shiny enclosure and UI?

      1. Didn´t mean to be negative. We both want the littlenones to be save emotionally and physicly. I´m a lazy hacker and not as innovative as you, otherwise i would have come up with it myself. Fake cry detection by AI sounds nice. My neuronal network is still fooled by that:)

  4. The A/V monitor we use does a really nice job cutting out the background noise and amplifying the higher pitched sounds. I’m guessing it is a fairly simple audio EQ+noise gate setup. But I’m kinda a KISS guy.

    1. This is an excellent point! My goal with this project was to explore an application of audio classification what could improve every day life, but this is also a great solution to the baby monitor problem itself! :)

  5. I fully understand that this is just a project and that [Ish Ot Jr.] is not using this as his actual baby monitor, but seriously why? This is not something anyone should actually want. I just don’t understand what the actual impetus is for this. Sure there are some noises that aren’t the baby that the monitor let’s through and they can be annoying. But filtering them out via software is a bad idea. Instead we should be looking into them (because it might be something bad) and eliminate them at the source. I know I sound like the super helicopter parent here, but even toying with the idea of delegating something as important as monitoring a baby to software that has the power to guess what is and is not important enough to alert you about is just not wise.

    Experiment as you will, but seriously please don’t use this in real life regardless of how confident you feel about it.

    1. people suggesting you should use this in real life:
      jack324: don’t use this in real life!!
      people suggesting you should use this in real life:
      jack324: I know you don’t exist, but still, don’t use this in real life!!!
      people suggesting you should use this in real life:

      Thanks for your concern. Instead of a baby monitor, let’s re-purpose it for detecting adults pointlessly whining – can you send me some audio samples so that I can re-train? 🙄

      1. I’m gonna end this here. Your idea is terrible. I’m not changing my opinion on that. Your not changing yours. Let’s both walk away from any further replies. Communicating further is only going to get both of us into a pointless argument. Have fun, good luck parenting.

        1. > Your not changing yours.

          Do trolls intentionally use poor grammar in order to elicit responses? 🤔

          Maybe keep walking – away from Hackaday, Hackster, and other sites like these where people share their ideas and fun projects with the community because they are passionate about hacking on stuff and creating new things. If you’re (FYI: “you are”) primarily interested in providing feedback on actual products, maybe check out https://www.producthunt.com/ or become a secret shopper. Happy trolling! 🙄

  6. This is a great project.

    Many years ago I bought one of these to baby sit a ceramic hot press that was constantly needing attention. Since I sat away from the lab, then I needed a way to get to the machine as soon as it had an alarm. These things go off with anything so it worked wonders.

    Now father of three, got rid of baby distress callers after number one. If babies need your attention, they’ll make sure they get it.

  7. To me, it seems like a good tutorial project with an interesting application.

    Nothing to be annoyed or to flame over; inadequate upbringing due to baby monitor with bad filters, perhaps?

  8. As a father of five I can relate to the sentiment that perhaps inspired this work but I think that for now it is ill advised to think that you can ignore your baby except for when a machine tells you there is something to notice. Raising children is hard work, some of that is unavoidable and not even humans get it right a lot of the time, there is no way around that until AI is superior to humans. I suggest that people considering parenthood give themselves the opportunity to experience what living with a infant is like and if you can’t manage the stress then perhaps get yourself desexed. I’m serious, let (un)natural selection solve this “problem”.

Leave a Reply

Please be kind and respectful to help make the comments section excellent. (Comment Policy)

This site uses Akismet to reduce spam. Learn how your comment data is processed.