Hacker News Reader

Comment by taeric | Hacker News Reader

Ironically this is a poster for complex protocol that seems easy. Consider the difficulties in distinguishing beeps.

There's important contextual information here. If I ask Alexa to tell me when the oven is preheated the next beep is probably going to be the oven.

If she's wrong it's not the end of the world. You don't need 100% accuracy for the solution to be useful. In fact you probably need to attend to whatever else was beeping anyways.

Maybe not the end of the world, but surely a fire hazard?

Probably only for a false negative, rather than a false positive; and surely the probability of the former can be minimised by cranking up the sensitivity (and not being picky about what counts as a beep).

An appliance's timer is not a safety feature. They are required to have their own fault detection and automaticity shutdown.

It's crappy UX.

1. You tell Alexa to let you know when the oven is preheated.

2. She hears the washing machine beep and tells you the oven is preheated.

3. You come back, find that the oven is still not ready.

4. You conclude it's probably the washing machine and go check it out, forgetting to tell Alexa to let you know when the oven is preheated.

It's the same UX I have with my visiting aunt. We've managed with it just fine for a long time.

It's a case of different expectations. You expect an "intelligent machine" to get it right and you're exasperated when it gets it wrong.

It's kind of like saying that using a rotary phone UI on your smartphone would have been okay, because we've managed with it just fine for a long time back when we had rotary phones.

Even in a limited form its still something new that we've never had before. For people that lived alone there was simply no other option but to be nearby to hear the beep. The whole point of Alexa is to have something that can take over some of the things another human would normally have helped you with.

Perfect is the enemy of good.

What if the audio pitches were outside human detection and distinct enough patterns?

Then it would work better, although you could still get false positives from neighbors' appliances when your windows and their windows are open. That, however, is completely beside the point.

The point is that the audio pitches aren't outside human detection and don't have distinct enough patterns right now. Changing pitch and patterns would require appliance manufacturers' support, at which point you might as well go with mature, well-defined and well-implemented solutions, such as WiFi connectivity, rather than defining and implementing your own protocol based on audio pitches and patterns.

I understand that, but wouldn't a pattern generator that selects from a series of pitches and a larger series of patterns produce sufficiently unique enough items to ensure that the probability of overlap to be virtually null?

It's just a thought, like I'm against the whole thing in general, but it seems like this is doable without needing a lot of support

If you simplify it to just "An appliance beeped in {room}" you avoid that problem as the laundry and kitchen would be different rooms at least, in many cases.

Often in Europe they are in the same place. YMMV

Sure not every home is the same, but I was particularly responding to the concern about running to the wrong room first.

To be clear. I was not trying to say this is not something that should or could be done. Just that it wasn't as simple as it sounded. The "simple protocol" of listening for the correct beep is actually fairly complicated.

A major amazon echo marketing component is seven directional microphones. It should be easy to stick an echo on the wall such that left is the washer and right is the dryer or whatever.

Also despite sounding similar to tone deaf humans one probably runs at 4001.342 Hz and the other at 3999.114 Hz which likely sounds the same to me, but to a computer is likely very obvious.

Sounds like a problem that would lend itself well to machine learning

I would think manually training your Echo would be easier than waiting for it to learn by amassing a large number of beeps (and trying to infer patterns from timestamps and your presence/absence). I would rather say something like "Alexa, listen to this notification sound, save it as 'oven done'." Then in future to be able to just say "Alexa, do VERB when 'oven done'".

Alternatively, Alexa could ask "I just heard a ding, was that your dishwasher?" and maybe compare against a database of known devices.

Not easy, but do you think it's more difficult than speech recognition?

And also the difficulties in distinguishing an appliance beep from what only sounds like an appliance beep, such as a truck reversing in front of your home.

I'd think a computer would be great at distinguishing unique beeps from each other. It'd just have to know which ones come from which named appliances, and that's something that could easily be crowdsourced.

Reply on news.ycombinator.com