WEBVTT - How Smart Speakers Work 0:00:04.240 --> 0:00:07.240 Welcome to Tech Stuff, a production of I Heart Radios 0:00:07.320 --> 0:00:14.000 How Stuff Works. Hey there, and welcome to tech Stuff. 0:00:14.040 --> 0:00:16.880 I'm your host, Jonathan Strickland. I'm an executive producer with 0:00:16.920 --> 0:00:19.960 I Heart Radio and I love all things tech, and guys, 0:00:19.960 --> 0:00:24.079 stick with me. I am fighting off a cold. You'll 0:00:24.120 --> 0:00:25.560 be able to hear it in my voice. I have 0:00:25.760 --> 0:00:28.720 no doubt. But you know, I wanted to get you 0:00:28.760 --> 0:00:32.040 guys a brand new episode. So we're gonna fight on 0:00:32.200 --> 0:00:37.200 because the show must keep going. I think I think 0:00:37.200 --> 0:00:40.120 this is saying, oh no, this cold medicine is good though. 0:00:40.200 --> 0:00:42.959 All right, Anyway, I thought that we would do an 0:00:42.960 --> 0:00:47.760 episode about smart speakers because I wanted to kind of 0:00:47.760 --> 0:00:51.199 start this whole episode off with with an old man observation, 0:00:51.240 --> 0:00:54.000 you know, get off my lawn kind of thing. And 0:00:54.080 --> 0:00:57.600 this is from our resident old man, old man Strickland. 0:00:57.840 --> 0:01:01.160 That meaning meaning me, So, when I was young, speakers 0:01:01.160 --> 0:01:03.959 were dumb. Now I don't. I don't mean that speakers 0:01:03.960 --> 0:01:07.479 were useless, or that they were terrible, or that they 0:01:07.480 --> 0:01:11.679 were incapable of replicating certain frequencies or volumes of sound, 0:01:12.319 --> 0:01:15.440 or that they were limited in some other way other 0:01:15.480 --> 0:01:19.319 than they didn't quote unquote think they didn't connect to 0:01:19.360 --> 0:01:22.800 any sort of computational engine in a meaningful way. You 0:01:22.880 --> 0:01:25.160 might have a set of speakers plugged into a computer, 0:01:25.600 --> 0:01:27.840 but that was just a one way communications tool, right. 0:01:27.840 --> 0:01:29.920 It was just a way to provide an outlet for 0:01:30.120 --> 0:01:33.200 sound that your computer was generating, nothing more than that. 0:01:33.880 --> 0:01:37.200 But contrast that with today, when we have numerous smart 0:01:37.240 --> 0:01:40.440 speakers on the market. These speakers act as a user 0:01:40.480 --> 0:01:44.840 interface between us and the Internet at large, often facilitated 0:01:44.840 --> 0:01:49.120 by a virtual assistant of some kind. Now with these speakers, 0:01:49.200 --> 0:01:52.760 we don't just listen to stuff like music and podcasts 0:01:52.840 --> 0:01:56.640 and the radio and you know, other traditional audio content. 0:01:57.200 --> 0:02:00.520 We use them to find out information. We might link 0:02:00.640 --> 0:02:03.360 them to our calendars so that we can get reminders 0:02:03.360 --> 0:02:06.760 for upcoming appointments. We probably use them to ask about 0:02:06.800 --> 0:02:09.720 the weather report. I use mine at home for that 0:02:09.800 --> 0:02:12.640 all the time, or even more often than that, if 0:02:12.639 --> 0:02:15.200 you're at my house, you'll hear us use it to 0:02:15.240 --> 0:02:17.400 find out which foods are safe for us to feed 0:02:17.480 --> 0:02:21.080 to our dog. My doggie, Tibolt, absolutely loves our smart 0:02:21.120 --> 0:02:24.160 speaker because it frequently gives us permission to spoil him 0:02:24.160 --> 0:02:27.560 with a carrot or a piece of banana. But how 0:02:27.639 --> 0:02:31.200 do these smart speakers work, How are they able to 0:02:31.320 --> 0:02:35.680 respond to our requests? And what are their limitations? How 0:02:35.760 --> 0:02:38.320 safe are they? That's the sort of stuff we're gonna 0:02:38.320 --> 0:02:40.920 be looking into in this episode of tech Stuff, and 0:02:40.960 --> 0:02:44.120 we'll start off with the basics, which means we have 0:02:44.160 --> 0:02:47.519 to start off with how speakers work in general. Now, 0:02:47.520 --> 0:02:49.840 this is something that I've covered before on tech Stuff, 0:02:50.120 --> 0:02:51.919 but I want to go over it again from a 0:02:52.000 --> 0:02:55.400 high level because well, I just find it fascinating that 0:02:55.480 --> 0:02:58.800 people figured out how to harness electricity to drive a 0:02:58.840 --> 0:03:02.760 motor so that it could in turn cause components to 0:03:02.800 --> 0:03:07.079 replicate a recorded or transmitted sound. And really motors being 0:03:07.120 --> 0:03:10.600 too generous, but to drive an element to create vibrations 0:03:10.639 --> 0:03:14.320 that could replicate a sound that was made into another component, 0:03:14.560 --> 0:03:17.080 that whole thing just boggles my mind that people are 0:03:17.120 --> 0:03:20.920 smart enough to figure that out. Okay, So to understand 0:03:20.960 --> 0:03:24.000 how speakers work, it first helps to understand how sound 0:03:24.160 --> 0:03:28.800 itself works. Sound is a physical phenomenon. Do do do do? 0:03:29.320 --> 0:03:33.560 Sound is all about vibrations, and typically we experience sound 0:03:33.600 --> 0:03:36.280 when we pick up on changes in air pressure that 0:03:36.520 --> 0:03:40.119 enter through our ear canal and then affect the tympanic 0:03:40.160 --> 0:03:44.800 membrane or ear drum. So it's all about these changes 0:03:44.880 --> 0:03:48.720 of of of air pressure, all about air molecules transmitting 0:03:48.800 --> 0:03:53.520 vibrations from a source outward in a radiating pattern from 0:03:53.520 --> 0:03:56.760 from that source. So let's think of someone knocking on 0:03:56.800 --> 0:03:59.520 a door. For example, you're inside a house, someone's knocking 0:03:59.560 --> 0:04:02.960 on your door. When that person's hand hits the door, 0:04:03.280 --> 0:04:07.600 it causes the door to vibrate, and that vibration transmits 0:04:07.640 --> 0:04:10.440 to the surrounding air molecules on the other side of 0:04:10.440 --> 0:04:13.640 the door. They get pushed through that vibration and then 0:04:13.640 --> 0:04:18.440 pulled when the the wood is vibrating back towards its 0:04:18.440 --> 0:04:23.359 original position. So the air molecules vibrate, those air molecules 0:04:23.400 --> 0:04:26.919 cause the next surrounding layer of air molecules to vibrate 0:04:26.960 --> 0:04:29.160 as well, and so on and so forth. It's like 0:04:29.200 --> 0:04:32.360 a cascade or domino effect. You get these little pockets 0:04:32.400 --> 0:04:35.599 of high and low air pressure that travel outward from 0:04:35.720 --> 0:04:40.919 that door. It spreads further as it goes towards you know, 0:04:41.000 --> 0:04:45.760 any distance, and if you are close enough so that 0:04:46.360 --> 0:04:49.200 you can still detect those changes in air pressure. You 0:04:49.360 --> 0:04:52.320 experience this by hearing the knocking on the door. Those 0:04:52.360 --> 0:04:56.039 vibrating air molecules lose a bit of energy as they 0:04:56.080 --> 0:04:59.200 move outward. Right, as they vibrate to the next layer, 0:04:59.320 --> 0:05:01.159 you start to lo use a bit of energy with 0:05:01.279 --> 0:05:05.560 each transmission of that So the sound gets quieter the 0:05:05.640 --> 0:05:08.800 further away you are because there's not as many air 0:05:08.839 --> 0:05:13.680 molecules vibrating, its amplitude as decreased. So if you are 0:05:13.720 --> 0:05:16.120 in hearing range, you can pick up on those changes 0:05:16.120 --> 0:05:18.720 of air pressure they encounter the tympanic membrane in your 0:05:18.720 --> 0:05:21.760 ear canal. Those changes in pressure will cause a reaction 0:05:21.800 --> 0:05:25.919 in your middle and inner ear set that will ultimately 0:05:25.920 --> 0:05:29.720 get picked up by your brain that interprets it as sound. Now, 0:05:29.720 --> 0:05:34.400 the frequency at which those fluctuations occur relate to the 0:05:34.440 --> 0:05:40.880 pitch that we hear, so faster vibrations are higher pitches, 0:05:41.080 --> 0:05:44.760 higher frequencies, higher notes. If you think of a musical scale, 0:05:45.520 --> 0:05:50.200 we perceive the force of the changes as volume, so 0:05:50.839 --> 0:05:54.560 lower forces lower volume right, and higher forces higher volume. 0:05:55.279 --> 0:05:58.039 The human ear can hear a pretty decent range of 0:05:58.080 --> 0:06:02.239 frequencies from twenty hurts, which means twenty cycles or twenty 0:06:02.360 --> 0:06:06.880 waves per second past a given point of reference, to 0:06:07.000 --> 0:06:12.320 twenty killer hurts. That's twenty thousand cycles or waves per second. 0:06:12.800 --> 0:06:15.440 So yeah, the cycle refers to the frequency of the 0:06:15.480 --> 0:06:19.120 wavelength of sound. The lower the frequency, the lower the sound. 0:06:19.440 --> 0:06:21.560 All right, and then our brain has to make meaning 0:06:21.560 --> 0:06:23.880 of all this, Right, it's not just that it's picking 0:06:23.960 --> 0:06:28.279 up on it. Our brain interprets this and we experience 0:06:28.360 --> 0:06:32.359 it as a sound we have heard. So it either 0:06:32.720 --> 0:06:36.840 matches this perceived sound with one we've encountered before, and 0:06:36.880 --> 0:06:39.840 then we say, oh, I know what that is. That's 0:06:39.880 --> 0:06:43.799 someone knocking at the door, or they might be Holy Cala, 0:06:44.000 --> 0:06:46.120 I've never heard that sound in my life. I have 0:06:46.200 --> 0:06:49.920 no idea what it is. If the sound is language, 0:06:50.000 --> 0:06:52.560 then our brains have to derive the meaning from the 0:06:52.640 --> 0:06:56.479 perceived sound. We've heard someone say words such as you're 0:06:56.520 --> 0:07:00.920 hearing me say this. Then our brains have to take 0:07:01.160 --> 0:07:03.960 that collection of sounds and say, what does that actually mean? 0:07:04.040 --> 0:07:07.200 What is the the context, what is the the intent? 0:07:07.640 --> 0:07:10.440 What is the message here? Otherwise it would just be 0:07:10.960 --> 0:07:14.760 you know, random noises that I'm making with my mouth. Alright, 0:07:14.800 --> 0:07:17.760 so we have a basic understanding behind the physics of sound. 0:07:17.760 --> 0:07:21.600 Now to talk about speakers and microphones and the reason 0:07:21.760 --> 0:07:24.000 I'm going to talk about both of them is that 0:07:24.080 --> 0:07:26.720 the devices complement one another. You can think of one 0:07:26.760 --> 0:07:31.080 as being the other in reverse. Plus smart speakers we 0:07:31.160 --> 0:07:34.160 have to talk about microphones anyway, because smart speakers have 0:07:34.400 --> 0:07:38.280 microphones as well as the speaker element. So you can 0:07:38.360 --> 0:07:41.520 think of this as one long process of taking the 0:07:41.520 --> 0:07:46.280 physical phenomena of sound waves, transforming that physical phenomena into 0:07:46.360 --> 0:07:49.800 an electrical signal, taking the electrical signal, and changing it 0:07:49.920 --> 0:07:52.920 back into something that can produce the sound waves that 0:07:53.040 --> 0:07:56.520 started the whole thing. So you're replicating the original sound 0:07:56.560 --> 0:08:00.480 waves with this end device, which in this case is 0:08:00.480 --> 0:08:03.120 allowed speaker. So the microphone is the part of the 0:08:03.160 --> 0:08:05.280 process where you take the sound and you turn it 0:08:05.320 --> 0:08:08.080 into an electrical signal, and the speakers where you take 0:08:08.120 --> 0:08:10.600 the electrical signal and you turn it back into actual sound. 0:08:10.680 --> 0:08:14.640 That's the simple way. But what's actually happening, Well, let's 0:08:14.680 --> 0:08:18.520 talk about on a physical level. Sound waves go into 0:08:18.560 --> 0:08:23.080 a microphone. So you've got these fluctuations and air pressure 0:08:23.200 --> 0:08:27.120 that encounter a microphone. I'm speaking into a microphone right now, 0:08:27.240 --> 0:08:30.480 so this is happening right now. Inside the microphone is 0:08:30.520 --> 0:08:33.679 a very thin diaphragm, typically made out of a very 0:08:33.720 --> 0:08:37.440 flexible plastic, and it's sort of like the skin of 0:08:37.440 --> 0:08:40.880 a drum. So as the changes in air pressure encounter 0:08:41.360 --> 0:08:45.520 the diaphragm, they cause the diaphragm to move back and forth. Well. 0:08:45.520 --> 0:08:49.319 Attached to the diaphragm is a coil of conductive wire, 0:08:49.760 --> 0:08:53.640 and that coil wraps either around or near a permanent magnet. 0:08:54.040 --> 0:08:57.200 Magnets have magnetic fields. They have a north pole and 0:08:57.240 --> 0:09:00.760 a south pole, and there's a magnetic field that surrounds 0:09:01.320 --> 0:09:05.720 the magnet. And the electro magnetic effect means that if 0:09:05.720 --> 0:09:10.600 you move a coil of conductive wire through a magnetic field, 0:09:11.040 --> 0:09:14.280 it will produce a change in voltage in that coil, 0:09:14.600 --> 0:09:19.000 otherwise known as electromotive force, and that means electrical current 0:09:19.040 --> 0:09:22.760 will flow through the coil. Now, if you have the 0:09:22.880 --> 0:09:26.240 end of that coil attached to a wire, a conductive 0:09:26.280 --> 0:09:30.120 wire for that current to flow through, you can send 0:09:30.160 --> 0:09:33.960 that current onto other components. So for our purposes, the 0:09:34.000 --> 0:09:37.360 component in question would be an amplifier, and I'll get 0:09:37.400 --> 0:09:40.480 to explaining why that is in just a moment, but 0:09:40.559 --> 0:09:43.160 first let's talk about loud speakers, and the way allowed 0:09:43.160 --> 0:09:48.000 speaker works is essentially the reverse of a microphone. You've 0:09:48.040 --> 0:09:51.440 got your permanent magnet around or near which is a 0:09:51.480 --> 0:09:56.360 coil of conductive wire. The wire is connected to a diaphragm, 0:09:56.400 --> 0:09:59.600 one much larger and typically made out of stiffer material 0:09:59.800 --> 0:10:03.480 that the plastic you'd find in a microphone. This is 0:10:03.520 --> 0:10:06.520 the element inside a speaker that will vibrate, that will 0:10:06.559 --> 0:10:10.960 push air and pull air as it moves either outward 0:10:11.040 --> 0:10:14.800 or inward. The electrical signal comes from a source such 0:10:14.880 --> 0:10:17.440 as the microphone we were just using a second ago 0:10:18.080 --> 0:10:22.439 that comes into the loudspeaker and it flows through the coil. Now, 0:10:22.480 --> 0:10:26.400 when you have an electrical current flowing through a conductive coil, 0:10:26.920 --> 0:10:31.079 you generate a magnetic field because the laws of electromagnetism. 0:10:31.600 --> 0:10:35.920 You've got the electro magnetic field generated as a result. 0:10:36.280 --> 0:10:39.440 Now that field will interact with the magnetic field of 0:10:39.480 --> 0:10:42.360 the permanent magnet. That the permnet magnet always has a 0:10:42.360 --> 0:10:46.040 magnetic field. The coil only has one when electric current 0:10:46.120 --> 0:10:48.760 is flowing through it. And as I said, we have 0:10:48.840 --> 0:10:51.120 magnets to have a north pole and a south pole. 0:10:51.160 --> 0:10:54.240 And we also know that when we bring two magnets 0:10:54.240 --> 0:10:57.840 with their north poles together, they'll push against each other, 0:10:57.960 --> 0:11:02.240 right because like repels like, But if we turn one 0:11:02.240 --> 0:11:04.640 of those magnets around so that now it's a south 0:11:04.679 --> 0:11:08.560 pole and a north pole, they attract one another, you know, 0:11:08.600 --> 0:11:15.160 opposites attract. So by having the this magnetic field being 0:11:15.200 --> 0:11:21.360 generated by the coil, uh, it starts to generate interactions 0:11:21.400 --> 0:11:25.520 with the magnetic field of the permanent magnet, so they 0:11:25.600 --> 0:11:28.160 start to push and pull against each other. Well, the 0:11:28.240 --> 0:11:31.959 coil is attached to that diaphragm, so it in turn 0:11:32.160 --> 0:11:36.000 drives the diaphragm to either push outward or pull inward. 0:11:36.480 --> 0:11:40.760 That causes air molecules to vibrate, just as it would 0:11:41.120 --> 0:11:43.840 with any other you know, source of sound, and it 0:11:43.920 --> 0:11:48.599 emanates outward from the loudspeaker, so you get a representation 0:11:48.920 --> 0:11:51.839 of the same sound that was going into the microphone 0:11:52.679 --> 0:11:56.760 got converted into an electrical current. The electrical current then 0:11:57.080 --> 0:12:00.360 was passed through a coil and next to a permanent 0:12:00.360 --> 0:12:03.720 magnet to create the same sort of movement. It replicates 0:12:03.720 --> 0:12:07.240 the movement of the original diaphragm in the microphone and 0:12:07.320 --> 0:12:11.200 generates the sound. So you get the replication of the 0:12:11.240 --> 0:12:15.079 sound that was made in the other location. It's pretty cool. 0:12:15.160 --> 0:12:18.400 I think now I did mention earlier that you would 0:12:18.400 --> 0:12:21.480 need an amplifier. And the reason you need an amplifier 0:12:21.640 --> 0:12:24.920 is that the electrical signal generated by a microphone is 0:12:24.960 --> 0:12:28.440 far too weak to drive allowed speakers diaphragm. You just 0:12:29.160 --> 0:12:31.880 wouldn't have the juice to do it. It would be 0:12:32.000 --> 0:12:35.839 much much less, uh powerful than what the speaker would need. 0:12:36.120 --> 0:12:39.040 So chances are the diaphragm would either not move at 0:12:39.080 --> 0:12:40.920 all because it would just be too stiff, it would 0:12:41.240 --> 0:12:44.559 resist the movement too much, or it would move so 0:12:44.600 --> 0:12:47.360 weakly as to generate little to no sound, so it 0:12:47.360 --> 0:12:50.559 wouldn't do you any good. So the signal from the 0:12:50.600 --> 0:12:53.240 microphone has to first pass through an amplifier, which, as 0:12:53.280 --> 0:12:56.679 the name implies, takes an incoming signal and increases the 0:12:56.720 --> 0:13:00.960 amplitude of that signal the volume. In other words, uh so, 0:13:01.000 --> 0:13:03.480 it doesn't affect pitch, but it does affect the signal 0:13:03.559 --> 0:13:08.160 strength and consequently the volume. And I've done episodes about amplifiers, 0:13:08.240 --> 0:13:11.920 including explaining the difference between amplifiers that use vacuum tubes 0:13:11.960 --> 0:13:14.880 and ones that use transistors, so I'm not going to 0:13:15.000 --> 0:13:18.640 go into that here. Besides, it doesn't really factor into 0:13:18.679 --> 0:13:22.679 our conversation about smart speakers anyway. It's just important for 0:13:23.000 --> 0:13:26.080 it to work with a microphone and speaker setting. Now, 0:13:26.120 --> 0:13:29.600 over the years, engineers have paired microphones and speakers and 0:13:29.720 --> 0:13:33.440 lots of stuff. You've got telephones, you've got intercom systems, 0:13:33.480 --> 0:13:37.280 public address systems, handheld radios, all sorts of things, so 0:13:37.320 --> 0:13:41.160 that technology was well and truly mature. Before we ever 0:13:41.240 --> 0:13:45.040 got our first smart speaker, there wasn't much call to 0:13:45.160 --> 0:13:49.200 incorporate microphones into home speaker systems for many years. I mean, 0:13:49.760 --> 0:13:52.560 what would you actually use a microphone embedded in a 0:13:52.640 --> 0:13:56.320 speaker for? Before smart speakers, Typically you would have your 0:13:56.360 --> 0:13:59.280 speakers like I'm talking about, like like sound system speakers. 0:13:59.400 --> 0:14:01.800 You would have them hooked up to some other dumb 0:14:02.080 --> 0:14:05.800 as in, not connected to a network technology. So it 0:14:05.880 --> 0:14:09.040 might be a sound system or home entertainment set up 0:14:09.040 --> 0:14:11.480 with a television as the focal point, or maybe even 0:14:11.720 --> 0:14:14.079 you know, a computer for the purposes of playing more 0:14:14.160 --> 0:14:19.240 dynamic sounds for like video games and and things like that. Um. 0:14:19.320 --> 0:14:21.800 But for a very long time, these were all thought 0:14:21.840 --> 0:14:25.320 of as one way communications applications, right, Like, the sound 0:14:25.400 --> 0:14:27.480 was coming from a source and it would get to 0:14:27.600 --> 0:14:30.800 us through the speakers, but we weren't meant to send 0:14:31.360 --> 0:14:34.480 sound back through those same channels. The information was just 0:14:34.560 --> 0:14:37.440 coming to you. You weren't sending anything back, But that 0:14:37.480 --> 0:14:40.040 would all change in time. Now. One thing to keep 0:14:40.040 --> 0:14:42.680 in mind about smart speakers is that they are the 0:14:42.680 --> 0:14:46.360 product of several different technologies and lines of innovation and 0:14:46.400 --> 0:14:50.800 development that all converged together. The microphone and speaker technology 0:14:51.120 --> 0:14:53.160 is one of the oldest ones that we can point 0:14:53.200 --> 0:14:57.000 to as far as the fundamental underlying technology is concerned, 0:14:57.560 --> 0:15:00.440 the stuff that's been around since the late nineties century. 0:15:00.600 --> 0:15:03.440 Now there is one other we'll talk about that's even older. 0:15:03.720 --> 0:15:06.440 But I don't want to spoil things. I'll just mention 0:15:06.920 --> 0:15:10.560 there is an even older line of development that goes 0:15:10.600 --> 0:15:14.240 into smart speakers than the microphone speaker stuff of the 0:15:14.320 --> 0:15:18.040 nineteenth century. Most of the other components, however, are much 0:15:18.080 --> 0:15:23.239 younger than that. One big one is speech or voice recognition. 0:15:23.600 --> 0:15:28.040 Creating computer systems that could detect noise was relatively simple. Right. 0:15:28.120 --> 0:15:31.120 You could have a computer connected to microphones and they 0:15:31.120 --> 0:15:35.360 could monitor the input from those microphones and any incoming 0:15:35.400 --> 0:15:38.680 signal could be registered. Right, they could record an incoming 0:15:38.720 --> 0:15:42.080 signal that would indicate the microphone had detected a noise. 0:15:42.560 --> 0:15:46.080 That's child's play. That's easy to do. But teaching computers 0:15:46.080 --> 0:15:49.160 how to analyze those signals and decipher them so that 0:15:49.160 --> 0:15:53.440 the computer could display in text or otherwise act upon 0:15:54.000 --> 0:15:57.880 that that sound in a meaningful way that was much 0:15:57.880 --> 0:16:02.400 more difficult. There was an IBM engineer named William C. 0:16:02.680 --> 0:16:06.560 Dirsh of the Advanced System Development Division who created an 0:16:06.640 --> 0:16:11.200 early implementation of voice recognition. It was a very limited application, 0:16:11.280 --> 0:16:14.240 but it proved that the ability to interact with computers 0:16:14.280 --> 0:16:18.800 by voice was more than just science fiction. Within IBM. 0:16:18.800 --> 0:16:23.080 It was called the Shoebox. Dirsh worked on this project 0:16:23.200 --> 0:16:26.440 in the early nineteen sixties and what he produced was 0:16:26.480 --> 0:16:29.840 a machine that had a microphone attached to it. The 0:16:29.880 --> 0:16:34.680 machine could detect sixteen spoken words, which included the digits 0:16:34.800 --> 0:16:39.160 of zero to nine plus some command indicators like plus 0:16:39.520 --> 0:16:43.360 minus total, sub total. You get the idea. So you 0:16:43.360 --> 0:16:46.680 could speak a string of numbers and then commands to 0:16:46.840 --> 0:16:49.920 this device, then ask it to total everything and it 0:16:49.960 --> 0:16:52.000 would do so. So it was more or less a 0:16:52.080 --> 0:16:58.000 basic calculator with some voice interpretation incorporated into it. Now 0:16:58.280 --> 0:17:02.000 there's a great newsreel piece about this shoebox. There's a 0:17:02.040 --> 0:17:05.040 demonstration of it, and it came out in nineteen one, 0:17:05.480 --> 0:17:08.480 and I love that newsreel because it has that great 0:17:08.560 --> 0:17:10.520 music you would hear in the background of those old 0:17:10.560 --> 0:17:14.560 industrial and business films. Anyway, there's also a helpful chart 0:17:14.840 --> 0:17:19.159 that hangs in the background of that video where Dersh 0:17:19.320 --> 0:17:22.439 is actually explaining how it works. You can see a 0:17:22.480 --> 0:17:25.919 little bit behind him what the what is actually being 0:17:25.960 --> 0:17:30.520 analyzed and uh he broke the words down into phonemes 0:17:30.560 --> 0:17:36.720 and syllables, so phonemes being specific sounds that make up words. So, 0:17:36.760 --> 0:17:40.679 for example, the digit one is a single syllable word 0:17:40.960 --> 0:17:43.520 with a vowel sound right at the front. But you 0:17:43.600 --> 0:17:48.200 also have the word eight that's another single syllable word 0:17:48.480 --> 0:17:51.040 as a vowel sound right at the front, but it's 0:17:51.359 --> 0:17:55.280 different from one phonetically in that eight also has a 0:17:55.359 --> 0:17:59.159 plosive and has that hard t at the end. So 0:17:59.200 --> 0:18:02.919 the shoebox was limited not just in what words it 0:18:02.960 --> 0:18:07.720 could recognize, but also the types of voices it could recognize. 0:18:07.880 --> 0:18:10.760 Get someone who has a different dialect or manner of speech, 0:18:10.760 --> 0:18:12.800 and the machine might not be able to understand them 0:18:12.800 --> 0:18:16.119 because they're not pronouncing the words the same way that 0:18:16.280 --> 0:18:20.560 drsh did. This would be a big challenge in speech 0:18:20.560 --> 0:18:24.240 recognition moving forward, and it's also an example of where 0:18:24.280 --> 0:18:28.480 we find bias creeping into technology. And it's not necessarily 0:18:28.520 --> 0:18:32.359 a conscious thing, but if you have people designing a 0:18:32.400 --> 0:18:36.520 system and they're designing it based off their own uh, 0:18:36.680 --> 0:18:41.280 you know, speech patterns, their own pronunciations, their own dialects, 0:18:41.800 --> 0:18:44.879 then it may be that the system they create works 0:18:44.960 --> 0:18:48.040 really well for them and less well for anyone who 0:18:48.160 --> 0:18:51.440 isn't them, And the further away you are from their 0:18:51.480 --> 0:18:56.200 manner of speaking, the more frustration you will encounter as 0:18:56.240 --> 0:18:59.719 you try to interact with that technology. That's an example 0:18:59.760 --> 0:19:03.200 of s and in fact, if you read the histories 0:19:03.320 --> 0:19:06.359 of speech recognition and as we'll get too later natural 0:19:06.640 --> 0:19:10.119 language processing, you'll see a lot of people say it 0:19:10.200 --> 0:19:13.119 works great if you happen to be a white man, 0:19:13.840 --> 0:19:17.880 because the manner of speech was being or the people 0:19:17.920 --> 0:19:21.000 who were designing it were primarily white men who were 0:19:21.760 --> 0:19:26.000 uh typically aiming for a a what is considered a 0:19:26.080 --> 0:19:31.840 non accented American dialect somewhere in you know, the Eastern 0:19:31.920 --> 0:19:35.439 seaboard side. But that meant that if you did have 0:19:35.520 --> 0:19:39.639 an accent or a dialect, or you had a different vernacular, 0:19:40.200 --> 0:19:43.240 that it was harder for the systems to actually understand 0:19:43.240 --> 0:19:46.399 what you were saying. That's an example of bias. Well. 0:19:46.760 --> 0:19:49.359 The general strategy was again to break up speech and 0:19:49.400 --> 0:19:52.560 too constituent sound units, you know, those phonemes, and then 0:19:52.600 --> 0:19:55.879 to susse out which words were being spoken based on 0:19:55.920 --> 0:19:59.880 those phonemes, and that was done by digitizing the voice train, 0:20:00.160 --> 0:20:04.159 forming it from sound into data that represented stuff like 0:20:04.240 --> 0:20:08.320 the sounds frequency or pitch, and then matching up specific 0:20:08.359 --> 0:20:12.199 signal signal signatures with specific phone nmes. So generally the 0:20:12.240 --> 0:20:14.919 idea was that the computer system would monitor incoming sound, 0:20:15.280 --> 0:20:18.919 convert the sound into digital data, compare that data that 0:20:19.000 --> 0:20:22.679 had received with information stored in a database, and effort 0:20:22.720 --> 0:20:26.199 to look for matches. Uh. The shoebox database was just 0:20:26.320 --> 0:20:29.280 sixteen words and size. Later ones would be much larger, 0:20:29.320 --> 0:20:33.399 but pretty quickly people realized this was not an efficient 0:20:33.480 --> 0:20:37.640 way of doing speech recognition because the bigger the vocabulary, 0:20:37.840 --> 0:20:40.040 the more work intens of it was to build out 0:20:40.080 --> 0:20:43.520 those databases. So it wasn't something that people thought would 0:20:43.520 --> 0:20:48.560 be sustainable for very large vocabularies. But the Shoebox marked 0:20:48.560 --> 0:20:50.680 the beginning of a serious effort to create machines that 0:20:50.720 --> 0:20:53.720 could accept audio cues as actual input, and as we'll see, 0:20:54.080 --> 0:20:57.760 that's one important component for these smart speaker systems. I've 0:20:57.800 --> 0:20:59.560 got a lot more to say, but before I get 0:20:59.600 --> 0:21:09.760 into the next part, let's take a quick break. Now, 0:21:09.800 --> 0:21:13.480 obviously we didn't jump right into full voice recognition right 0:21:13.520 --> 0:21:17.520 after IBM S Shoebus innovation. The challenges related to building 0:21:17.560 --> 0:21:21.399 automated speech recognition systems were numerous, even for just a 0:21:21.520 --> 0:21:24.879 single language, because, as I said, you can have accents 0:21:24.960 --> 0:21:28.280 and dialects. One voice can have a very different tonal 0:21:28.400 --> 0:21:32.679 quality from another, people speak at different speeds. Teaching machines 0:21:32.720 --> 0:21:35.480 how to recognize speech when the phonemes and pacing of 0:21:35.520 --> 0:21:40.840 that speech aren't consistent from speaker to speaker, that's really hard. 0:21:41.320 --> 0:21:43.119 This kind of gets back to the same sort of 0:21:43.200 --> 0:21:46.680 challenges you have when you're teaching machines how to recognize images. 0:21:47.440 --> 0:21:51.080 You know, you teach a human what a coffee mug is. 0:21:51.119 --> 0:21:53.320 I always use this example, but you teach a human 0:21:53.359 --> 0:21:55.800 what a coffee mug is, and pretty soon they can 0:21:55.840 --> 0:22:00.000 extrapolate from that example and understand that coffee mugs can 0:22:00.000 --> 0:22:03.879 them in all different sizes and colors, and you know 0:22:04.240 --> 0:22:08.320 different designs and textures. We get it. Like you you 0:22:08.359 --> 0:22:11.640 see a couple of coffee mugs, you understand machines though 0:22:12.480 --> 0:22:15.280 they aren't able to do that. Machines, you know, you 0:22:15.320 --> 0:22:17.440 have to give them lots and lots and lots of 0:22:17.480 --> 0:22:20.479 different examples before they can start to pick up on 0:22:20.600 --> 0:22:24.960 what things actually make a coffee mug. Same sort of 0:22:25.000 --> 0:22:28.639 thing with speech, right, So if you don't have consistency 0:22:28.760 --> 0:22:31.679 between speakers, it makes it very hard for machines to 0:22:31.800 --> 0:22:34.800 learn what people are saying. Now, it didn't take long 0:22:34.880 --> 0:22:37.399 for the tech industry at large to really dive into 0:22:37.400 --> 0:22:41.520 trying to solve this problem. In ninete, DARPA, that's the 0:22:41.640 --> 0:22:45.359 Research and Development division of the United States Department of Defense, 0:22:45.760 --> 0:22:48.800 got behind speech recognition in a big way. Now, remember 0:22:49.280 --> 0:22:54.080 darp it self doesn't do research. The organization's purpose is 0:22:54.080 --> 0:22:58.280 to invite organizations to pitch projects that align with whatever 0:22:58.359 --> 0:23:01.879 darpest goals are and and DARBA would provide funding to 0:23:02.440 --> 0:23:07.000 the winning organizations to see these projects to completion if possible. 0:23:07.440 --> 0:23:09.840 So DARK is really more of a vetting and funding 0:23:10.000 --> 0:23:15.400 organization anyway. In n DARPA created a five year program 0:23:15.520 --> 0:23:20.160 called Speech Understanding Research or s u are. The initial 0:23:20.240 --> 0:23:23.320 goal was pretty darn ambitious considering the capabilities of the 0:23:23.359 --> 0:23:27.240 technology at the time. The project director, Larry Roberts, wanted 0:23:27.240 --> 0:23:30.440 a system that would be capable of recognizing a vocabulary 0:23:30.560 --> 0:23:34.119 of ten thousand words with less than ten percent error. 0:23:34.560 --> 0:23:37.240 After holding a few meetings with some of the leading 0:23:37.320 --> 0:23:41.840 computer engineers of the day, Roberts suggusted that goal significantly. 0:23:42.560 --> 0:23:45.359 After that adjustment, the target was going to be a 0:23:45.400 --> 0:23:50.040 system capable of recognizing one thousand words, not ten thousand. 0:23:50.920 --> 0:23:53.359 Nearror levels still had to be less than ten percent, 0:23:53.840 --> 0:23:55.720 and the goal was for the system to be able 0:23:55.760 --> 0:24:02.359 to accept continuous speech, as opposed to very deliberate speech 0:24:03.080 --> 0:24:08.000 with pauses between each pair of words that would not 0:24:08.119 --> 0:24:13.040 be really that useful. One person who was skeptical about 0:24:13.080 --> 0:24:16.760 the potential success of this project was John R. Pierce 0:24:16.960 --> 0:24:20.639 of Bell Labs. He argued that any success would be 0:24:20.720 --> 0:24:25.440 limited so long as machines remained incapable of understanding the words, 0:24:25.840 --> 0:24:28.720 not just recognizing a word based on phone names, but 0:24:28.840 --> 0:24:31.359 understanding what the word is. That is. Pierce felt that 0:24:31.359 --> 0:24:34.080 the machines needed some way to parse the language to 0:24:34.119 --> 0:24:37.040 get to the meaning of what was being said. That's 0:24:37.080 --> 0:24:38.919 an important idea that we will come back to in 0:24:38.960 --> 0:24:41.919 just a bit now. Among the companies and organizations that 0:24:42.040 --> 0:24:46.600 landed contracts with DARPA were a Carnegie Melon University BBN, 0:24:46.600 --> 0:24:49.080 which actually played a big part in developing our ponette, 0:24:49.080 --> 0:24:53.240 the predecessor to the Internet, Lincoln Laboratory, and several more 0:24:53.720 --> 0:24:56.840 and very smart people began to create systems intended to 0:24:56.880 --> 0:25:00.520 recognize speech and meaningful ways. The names of the programs 0:25:00.520 --> 0:25:02.800 were a lot of fun. There was h W I 0:25:03.119 --> 0:25:06.280 M that stood for hear what I mean as in 0:25:06.440 --> 0:25:09.040 here as in listen hear what I mean. That one 0:25:09.160 --> 0:25:14.320 was from BBN. CMU introduced hearsay, which was later designated 0:25:14.320 --> 0:25:17.080 as Hearsay one, and then they came out with Hearsay two. 0:25:17.560 --> 0:25:22.800 They also would demonstrate another one called harpy. Oh, and 0:25:22.840 --> 0:25:25.679 there was a professor at CMU named Dr James Baker 0:25:25.840 --> 0:25:29.439 who would design a system called Dragon in nineteen seventy 0:25:29.520 --> 0:25:32.040 five that he would later leverage into a company with 0:25:32.119 --> 0:25:35.480 his wife, Dr Janet M. Baker in the nineteen eighties, 0:25:35.520 --> 0:25:40.000 and they had a very successful business with speech recognition software. Now, 0:25:40.040 --> 0:25:42.399 I'm not going to go into each of those programs 0:25:42.400 --> 0:25:45.240 in deep detail, but rather just mentioned that they all 0:25:45.280 --> 0:25:48.879 helped advance the cause of creating systems that can recognize speech. 0:25:49.440 --> 0:25:51.480 One of the big developments that came out of all 0:25:51.520 --> 0:25:55.280 that work was a shift to probabilistic models, which would 0:25:55.320 --> 0:25:58.080 also play a really important part in another phase of 0:25:58.160 --> 0:26:00.680 developing the smart speaker. So what do I mean when 0:26:00.720 --> 0:26:04.520 I say probabilistic? Well, as the name indicates, it all 0:26:04.520 --> 0:26:08.760 has to do with probabilities. Essentially, systems would analyze incoming 0:26:08.760 --> 0:26:12.399 phonemes and make guesses as to what was being said 0:26:12.680 --> 0:26:15.520 based on the probability of it being a given word 0:26:15.720 --> 0:26:19.159 or part of a word. The systems typically go with 0:26:19.240 --> 0:26:22.920 whatever word has the highest probability of being the correct one. 0:26:23.640 --> 0:26:26.840 Even with that approach, there are nuances to language that 0:26:26.880 --> 0:26:29.840 are difficult to account for with a machine. So, for example, 0:26:29.840 --> 0:26:32.280 you have homonyms and which you have two words that 0:26:32.440 --> 0:26:35.720 sound the same but have very different meanings and potentially 0:26:35.760 --> 0:26:39.919 spellings like right as in to write a sentence, or 0:26:40.080 --> 0:26:43.040 right as in am I right? Or am I wrong? 0:26:43.600 --> 0:26:46.439 Or you could have a pair of words that sound 0:26:46.480 --> 0:26:49.560 like a single word and have confusion there, such as 0:26:49.880 --> 0:26:52.720 a door. You can say a door you mean you're 0:26:52.720 --> 0:26:55.840 meaning a single door a door to go into a building, 0:26:56.040 --> 0:26:58.520 or you might say a dore as an I adore 0:26:58.960 --> 0:27:02.040 this podcast you're doing, Jonathan. That's sweet of you, Thank 0:27:02.040 --> 0:27:06.399 you for saying that. So computer scientists were hard at 0:27:06.400 --> 0:27:10.080 work advancing both the capability of machines to make correct 0:27:10.200 --> 0:27:13.720 guesses at individual phone names and then full words, as 0:27:13.720 --> 0:27:15.920 well as figuring out a way to teach machines to 0:27:16.000 --> 0:27:20.959 adjust guesses based on context. That requires a deeper understanding 0:27:21.000 --> 0:27:24.520 of the language within which you're working. If you're aware 0:27:24.560 --> 0:27:27.439 of certain idioms, you can make a good guess at 0:27:27.480 --> 0:27:29.320 a word or phrase even if you didn't get a 0:27:29.320 --> 0:27:33.440 clean pass at it right. So, for example, the phrase 0:27:33.760 --> 0:27:37.200 it's raining cats and dogs just means it's raining a lot. 0:27:37.520 --> 0:27:40.359 And if a system included a database that indicated the 0:27:40.480 --> 0:27:44.760 phrase cats and dogs sometimes follows the phrase it's raining, 0:27:45.320 --> 0:27:47.640 then the system is more likely to guess the correct 0:27:48.280 --> 0:27:52.760 sequence of words instead of guessing something that sounded similar 0:27:52.840 --> 0:27:55.560 but it's wrong. For example, if it said, oh, they 0:27:55.600 --> 0:27:59.800 must have said it's raining bats and hogs, that would 0:27:59.800 --> 0:28:04.760 not makes sense. So the systems estimate the probability that 0:28:04.840 --> 0:28:08.359 any given sequence of sounds within the database matches what 0:28:08.480 --> 0:28:12.120 the systems have just quote unquote heard progress in this 0:28:12.160 --> 0:28:15.040 area was steady, but slow, and I'd argue that it 0:28:15.080 --> 0:28:17.959 was also a reminder that concepts like Moore's law do 0:28:18.040 --> 0:28:22.760 not apply universally across technology. Rapid development in one particular 0:28:22.800 --> 0:28:26.000 domain of technology is not necessarily an indicator that the 0:28:26.040 --> 0:28:28.760 same sort of progress will be observed in all other 0:28:28.840 --> 0:28:33.919 areas of tech. We often get into the mistaken habit 0:28:34.000 --> 0:28:37.200 of believing that Moore's law applies to everything. Alright. So 0:28:37.640 --> 0:28:42.120 a related concept to voice recognition is something called natural 0:28:42.240 --> 0:28:45.480 language processing, and this relates back to how we humans 0:28:45.480 --> 0:28:49.000 tend to process information compared to the way machines tend 0:28:49.040 --> 0:28:52.200 to do it. So we humans formulate ideas, we shape 0:28:52.200 --> 0:28:55.680 those ideas into words and sentences. We communicate them in 0:28:55.760 --> 0:28:59.239 some way to other people through that language. It may 0:28:59.280 --> 0:29:02.160 be through speed you maybe through text. It may even 0:29:02.160 --> 0:29:06.280 be through a nonverbal or non literary way, but we 0:29:06.400 --> 0:29:11.479 communicate those ideas. Machines typically accept input, they perform some 0:29:11.560 --> 0:29:15.400 process or sequence of processes on that input, and then 0:29:15.400 --> 0:29:18.960 they supply an output of some sort. Machines do this 0:29:19.040 --> 0:29:22.480 in machine language. That's a code that's far too difficult 0:29:22.480 --> 0:29:26.200 for humans to process. Easily. Binary is an example of 0:29:26.280 --> 0:29:30.520 machine language. Binary is represented as zeros and ones, which 0:29:30.520 --> 0:29:33.120 would group together can represent all sorts of stuff. But 0:29:33.160 --> 0:29:35.600 if you just looked at a big block of zeros 0:29:35.600 --> 0:29:38.280 and ones, it would mean nothing to you. It's not 0:29:38.360 --> 0:29:41.520 easy for humans to use, and then machines in turn 0:29:41.600 --> 0:29:44.960 are not natively able to understand human language, so there's 0:29:44.960 --> 0:29:49.720 a language barrier there. Because of that, people created different 0:29:49.760 --> 0:29:54.480 programming languages. These languages provide layers of abstraction from the 0:29:54.560 --> 0:29:57.960 machine language. They make it easier to create programs or 0:29:58.240 --> 0:30:01.560 directions that the computer should fall low. So the person 0:30:01.680 --> 0:30:04.640 who's doing the programming is using a programming language that's 0:30:04.640 --> 0:30:08.040 easy for humans to use that then gets converted into 0:30:08.240 --> 0:30:11.520 machine language that the computers understand. But what if you 0:30:11.560 --> 0:30:14.800 could send commands to a computer using natural language, not 0:30:14.880 --> 0:30:20.320 even programming language. You could just speak in Plaine vernacular, 0:30:20.400 --> 0:30:23.960 whether it's English or any other language, the way humans 0:30:24.000 --> 0:30:27.320 communicate with one another. What if a computer could extract 0:30:27.400 --> 0:30:30.760 meaning from a sentence, understand what it was you wanted 0:30:30.800 --> 0:30:34.280 the computer to do, and then respond appropriately. So imagine 0:30:34.280 --> 0:30:35.960 how much time you could save if you could just 0:30:36.040 --> 0:30:38.640 tell your computer what you wanted it to do, and 0:30:38.680 --> 0:30:41.440 it took care of the rest. If you had a 0:30:41.480 --> 0:30:46.280 powerful enough computer system with strong enough AI, maybe you 0:30:46.280 --> 0:30:49.480 could even potentially do something like describe a game that 0:30:49.560 --> 0:30:52.240 you would love to be able to play, like not 0:30:52.240 --> 0:30:54.400 not a game that exists, a game in your head, 0:30:54.880 --> 0:30:56.760 and you could describe it to a computer and the 0:30:56.800 --> 0:31:00.480 computer could actually program that game. Well, we're we're definitely 0:31:00.520 --> 0:31:03.480 not anywhere close to that yet, but we made enormous 0:31:03.520 --> 0:31:07.240 progress with natural language processing. Now, the history of natural 0:31:07.320 --> 0:31:11.760 language processing isn't exactly an extension of voice recognition. It's 0:31:11.760 --> 0:31:16.200 actually more like a parallel line of investigation. And that's 0:31:16.200 --> 0:31:20.760 because natural language processing doesn't require voice recognition. You can 0:31:20.840 --> 0:31:24.080 have an implementation in which you just right commands in 0:31:24.200 --> 0:31:26.440 natural language, you know, you type them out on a 0:31:26.520 --> 0:31:30.760 keyboard and the machine then carries out those those instructions. 0:31:30.800 --> 0:31:33.400 So much of the early work in natural language processing 0:31:33.440 --> 0:31:37.400 was in text based communication rather than in speech. The 0:31:37.480 --> 0:31:41.240 history of natural language processing includes stuff like the Turing test, 0:31:41.480 --> 0:31:44.840 named after Alan Turing. So the most common interpretation of 0:31:44.880 --> 0:31:47.560 the Turing test these days is that you've got a 0:31:47.600 --> 0:31:50.280 scenario in which a person is alone in a room 0:31:50.320 --> 0:31:53.080 with a computer terminal, they can type whatever they like 0:31:53.280 --> 0:31:57.520 into the computer terminal, and someone or something is responding 0:31:57.560 --> 0:32:00.719 to them in real time. Now it might be another person, 0:32:01.280 --> 0:32:04.120 or it might be a computer system that's responding to 0:32:04.360 --> 0:32:08.959 that person. You run a whole bunch of test subjects 0:32:08.960 --> 0:32:12.040 through this process, and if the computer system is able 0:32:12.080 --> 0:32:15.280 to fool a certain percentage of those test subjects, like 0:32:15.440 --> 0:32:18.720 say thirty percent of them, that it is in fact 0:32:18.760 --> 0:32:21.560 another human and not a computer, it is said to 0:32:21.640 --> 0:32:24.920 have passed the Turing test, And typically we use that 0:32:24.960 --> 0:32:27.400 to mean the machine has given off the appearance of 0:32:27.440 --> 0:32:31.680 possessing intelligence similar to the one that we humans possess. 0:32:32.400 --> 0:32:35.520 That gets beyond our scope for this episode, but it 0:32:35.560 --> 0:32:38.760 helps point out that stuff like speech recognition and natural 0:32:38.840 --> 0:32:42.280 language processing are both closely related to the field of 0:32:42.360 --> 0:32:45.720 artificial intelligence. In fact, they really belong within the artificial 0:32:45.760 --> 0:32:50.320 intelligence domain. The Turing test was more of a hypothetical. 0:32:50.560 --> 0:32:52.800 It was a bit of a cheeky way of saying, Hey, 0:32:53.000 --> 0:32:55.680 if you can't tell whether or not something is intelligent, 0:32:56.000 --> 0:32:58.560 it makes sense to treat it as if it actually 0:32:58.840 --> 0:33:02.520 is intelligent. After all, we assume that every human with 0:33:02.560 --> 0:33:06.440 whom we interact possesses some level of intelligence. Based on 0:33:06.520 --> 0:33:09.640 those interactions, so why should we not extend the same 0:33:09.680 --> 0:33:14.480 courtesy to machines. Now, natural language processing would prove to 0:33:14.480 --> 0:33:18.000 be another super challenging problem to solve. In computer science. 0:33:18.280 --> 0:33:21.880 Early work was done in translation algorithms, and these were 0:33:21.880 --> 0:33:24.800 programs that attempted to take phrases written in one language 0:33:24.840 --> 0:33:29.120 and translate those automatically into a second language. At first, 0:33:29.160 --> 0:33:33.960 that seemed pretty straightforward, but you realize that's also pretty tricky. Really. 0:33:34.200 --> 0:33:37.000 For one thing, you can't just translate word for word 0:33:37.160 --> 0:33:40.080 and keep the same order from one language to another. 0:33:40.520 --> 0:33:44.800 The syntax or the rules that the language follow uh, 0:33:44.840 --> 0:33:47.719 they could be different from language to language. In one language, 0:33:47.720 --> 0:33:50.960 you might use an infinitive such as to record, in 0:33:51.000 --> 0:33:53.760 the middle of a sentence, while another language might put 0:33:53.800 --> 0:33:56.400 all the infinitives at the end of a sentence. So 0:33:56.960 --> 0:33:59.240 in one language, I might say I'm going to record 0:33:59.280 --> 0:34:02.320 a podcast in the studio right now, but in another 0:34:02.400 --> 0:34:05.080 language it might come out as I'm going a podcast 0:34:05.080 --> 0:34:08.000 in the studio right now to record. It starts to 0:34:08.000 --> 0:34:12.879 sound like yoda. There was initial excitement around machine translation, 0:34:13.160 --> 0:34:16.400 but once computer scientists and linguists began to see the 0:34:16.440 --> 0:34:20.320 scope of this challenge, their excitement faded a bit. Also, 0:34:20.440 --> 0:34:22.400 there was a lot of other stuff going on in 0:34:22.440 --> 0:34:25.200 the nineteen sixties and seventies that was demanding a lot 0:34:25.200 --> 0:34:28.520 of attention, such as the Space race. So for a while, 0:34:28.880 --> 0:34:32.799 this branch of computer science was given less attention than 0:34:32.840 --> 0:34:37.160 other branches, and by less attention, I really mean funding. Now, 0:34:37.160 --> 0:34:39.359 when we come back, we'll talk a bit more about 0:34:39.400 --> 0:34:42.799 the advances that were necessary to support natural language processing, 0:34:43.000 --> 0:34:44.880 and we'll move on to how this would be another 0:34:44.960 --> 0:34:48.880 important component in smart speakers. But first, let's take another 0:34:49.000 --> 0:35:00.960 quick break. Okay, So early enthusiasm for an natural language 0:35:00.960 --> 0:35:04.520 processing created a bit of a hype cycle that ultimately 0:35:04.640 --> 0:35:10.160 crashed into the telephone poll of unmet expectations. That was 0:35:10.200 --> 0:35:15.280 a really bad metaphor. Anyway, natural language processing went through 0:35:15.520 --> 0:35:18.520 something similar to what we saw with virtual reality in 0:35:18.520 --> 0:35:22.920 the nineteen nineties. You know, people saw what was actually achievable, 0:35:23.480 --> 0:35:26.440 and then they compared that to what they thought they 0:35:26.440 --> 0:35:29.120 were going to get, and those two things didn't match 0:35:29.200 --> 0:35:31.919 up at all, and that really pulled the rug out 0:35:31.960 --> 0:35:35.440 of funding for natural language processing, which men of course, 0:35:35.480 --> 0:35:40.040 that progress slowed way down. It kept going, but it 0:35:40.239 --> 0:35:43.239 was definitely on the back burner for a lot of projects. 0:35:43.680 --> 0:35:46.799 When interest renewed in the nineteen eighties, there had been 0:35:46.800 --> 0:35:51.440 a shift in thinking around natural language processing. Computer scientists 0:35:51.480 --> 0:35:54.640 were starting to look at statistical approaches similar to what 0:35:54.719 --> 0:35:58.920 was going on with speech recognition, building up probabilistic models 0:35:58.960 --> 0:36:01.520 in which a computer can start making what amounts to 0:36:01.840 --> 0:36:05.880 educated guesses at the meaning of a command or a phrase. 0:36:06.480 --> 0:36:10.080 Machine learning became an important component on the back end 0:36:10.120 --> 0:36:13.839 of these systems, and later artificial neural networks became an 0:36:13.880 --> 0:36:17.719 important part as well. A neural network processes information in 0:36:17.760 --> 0:36:20.400 a way that's sort of analogous to how our brains 0:36:20.480 --> 0:36:24.719 do it. You have nodes or neurons that connect to 0:36:24.800 --> 0:36:29.000 other nodes, and each node affects incoming data in a 0:36:29.040 --> 0:36:32.279 certain way, performing some sort of operation on it, and 0:36:32.400 --> 0:36:34.960 the degree to which they do that in one way 0:36:35.080 --> 0:36:39.120 versus another is called the weight of that node. Computer 0:36:39.160 --> 0:36:42.799 scientists apply weights across the nodes in an effort to 0:36:42.880 --> 0:36:46.840 get a specific result in order to train these models. 0:36:46.880 --> 0:36:50.640 So you might feed a specific command into such a system, 0:36:50.680 --> 0:36:53.560 and you let it go through the computational process from 0:36:53.600 --> 0:36:56.080 the beginning of the neural network through to the end, 0:36:56.560 --> 0:36:58.520 and then you look at the result, and if the 0:36:58.560 --> 0:37:01.520 result is correct, well, that just means the system is 0:37:01.520 --> 0:37:04.719 already working as you intended it, which honestly is not 0:37:04.880 --> 0:37:08.480 likely to happen early on. But if it's not correct, 0:37:08.760 --> 0:37:12.400 then you start adjusting the weights on those nodes in 0:37:12.520 --> 0:37:15.200 order to affect the outcome. I almost think of it 0:37:15.239 --> 0:37:18.440 as like Plinko or pachinko, where you've got the little 0:37:18.520 --> 0:37:20.920 coin and you drop it down and it bounces on 0:37:20.920 --> 0:37:24.080 all the pegs and sometimes you're like you might think, 0:37:24.080 --> 0:37:25.520 all right, well, this time it's going to go right 0:37:25.520 --> 0:37:28.600 for that center slot, but it doesn't, and you think, well, 0:37:28.600 --> 0:37:31.080 maybe if I remove some of these pegs or I 0:37:31.239 --> 0:37:33.839 shift these pegs over a little bit, I can drop 0:37:33.880 --> 0:37:35.880 it in that same spot and get hit the center. 0:37:36.120 --> 0:37:38.160 It's kind of like that, except you're talking about data, 0:37:38.280 --> 0:37:42.120 not physical moving parts. So you have to do this 0:37:42.160 --> 0:37:46.960 a lot, like up to like millions of times in 0:37:47.080 --> 0:37:49.680 order to try and train a system so that responds 0:37:49.680 --> 0:37:53.399 appropriately to commands. And once it's trained, you can then 0:37:53.480 --> 0:37:55.799 test new commands on the system to see if it 0:37:55.800 --> 0:37:58.640 can parse them and respond appropriately. And in this way, 0:37:58.840 --> 0:38:03.240 the system quote unquote learns over time how to respond 0:38:03.440 --> 0:38:07.040 to commands. And then we have another component that's important 0:38:07.040 --> 0:38:10.839 with smart speakers, and that's speech generation. So it's one 0:38:10.840 --> 0:38:14.600 thing to have a machine either broadcast or play back 0:38:14.640 --> 0:38:18.080 a recording of speech. It's another thing for a machine 0:38:18.120 --> 0:38:22.319 to generate brand new speech. In computer science, we call 0:38:22.360 --> 0:38:26.919 it speech synthesis. Now, this is the really old technology 0:38:26.960 --> 0:38:29.279 I was alluding to at the beginning of this episode, 0:38:29.520 --> 0:38:32.960 speech synthesis. If you want to be really, you know, 0:38:33.040 --> 0:38:36.479 kind of technical about it, it actually predates every other 0:38:36.520 --> 0:38:39.759 technology I've mentioned up to this point, at least in 0:38:40.000 --> 0:38:43.880 its most rudimentary implementations. You have to go way back 0:38:43.960 --> 0:38:47.360 to the eighteenth century the seventeen seventies, as when a 0:38:47.440 --> 0:38:52.480 Russian smarty pants named Christian Kradsenstein was building a device 0:38:52.560 --> 0:38:56.800 that used acoustic resonators. These these reads that would vibrate, 0:38:57.160 --> 0:39:01.879 and it was in an attempt to replicate basic vowel sounds. Now, 0:39:01.920 --> 0:39:04.399 even with such a working device, it would be really 0:39:04.400 --> 0:39:07.960 difficult to communicate anything meaningful unless you were, i guess, 0:39:08.000 --> 0:39:11.399 speaking whale like Dory and finding Nemo. But it would 0:39:11.400 --> 0:39:13.640 be an early example of how people tried to create 0:39:13.680 --> 0:39:18.080 mechanical systems that could replicate speech or elements of speech. 0:39:18.440 --> 0:39:23.600 Another inventor named Wolfgang von Kimberland built an acoustic mechanical 0:39:23.719 --> 0:39:28.080 speech machine and that used reads and tubes and a 0:39:28.160 --> 0:39:31.520 pressure chamber, and it was all meant to replicate various 0:39:31.600 --> 0:39:35.640 speech sounds. He had other elements to create sounds like plosives, 0:39:35.640 --> 0:39:39.880 those hard sounds that I mentioned earlier in the episode. 0:39:40.520 --> 0:39:43.080 So he had all these different elements that, working together, 0:39:43.160 --> 0:39:47.640 could create parts of the sounds that we humans make 0:39:47.680 --> 0:39:51.480 when we speak. He also built a supposed chess playing machine, 0:39:51.920 --> 0:39:53.799 and it turned out that the chess playing part was 0:39:53.840 --> 0:39:58.680 a hoax. So unfortunately, because that device was a hoax, 0:39:58.840 --> 0:40:03.360 a lot of people dismiss his other work, which was legitimate. 0:40:03.880 --> 0:40:07.840 So by fudging on one thing, he kind of cast 0:40:08.040 --> 0:40:12.120 doubt on everything he had ever done. Skipping ahead quite 0:40:12.160 --> 0:40:15.720 a bit, we get to Homer Dudley, which is a 0:40:15.760 --> 0:40:21.640 fantastic name. He unveiled the voter or voice Operating Demonstrator 0:40:21.880 --> 0:40:25.480 device at the New York World's Fair in nineteen thirty nine. 0:40:25.960 --> 0:40:29.880 It consisted of a complex series of controls and it 0:40:29.960 --> 0:40:32.720 sort of reminds me of something like a musical instrument, 0:40:32.800 --> 0:40:36.840 kind of like a synthesizer, but with extra controlling units. 0:40:36.880 --> 0:40:39.400 Like there was like a wrist element, there was a pedal. 0:40:39.719 --> 0:40:43.080 There's a lot of stuff that made it very complex, 0:40:43.440 --> 0:40:47.239 and with a lot of practice, you could create specific 0:40:47.320 --> 0:40:51.360 sounds from this synthesizer. You could even create words or 0:40:51.440 --> 0:40:55.120 full sentences, though from what I understand, it was incredibly 0:40:55.200 --> 0:40:57.760 challenging to do. It was a very high learning curve, 0:40:58.120 --> 0:41:02.040 but it demonstrate the possibility of a like tronic synthesized speech. Now. 0:41:02.080 --> 0:41:06.120 There was a lot of work done in this field 0:41:07.120 --> 0:41:12.040 by lots of different talented scientists and engineers, and someday 0:41:12.080 --> 0:41:14.320 I'll have to do a full episode on the history 0:41:14.360 --> 0:41:17.839 of speech synthesis. It's really fascinating, but it's far too 0:41:17.920 --> 0:41:21.279 big a topic to cover in its entirety in this episode. 0:41:21.640 --> 0:41:24.480 By the late nineteen sixties we had our first text 0:41:24.640 --> 0:41:27.920 to speech system, and by the late nineteen seventies and 0:41:28.000 --> 0:41:31.280 early nineteen eighties, the state of the art had progressed 0:41:31.360 --> 0:41:33.160 quite a bit and we were starting to get to 0:41:33.200 --> 0:41:38.360 a point where we could create very understandable computer voices. 0:41:38.400 --> 0:41:41.680 They weren't natural, they didn't sound like people, but you 0:41:41.719 --> 0:41:45.439 could understand what they were saying. And finally, something else 0:41:45.440 --> 0:41:48.840 that would enable smart speakers and virtual assistance was the 0:41:48.880 --> 0:41:53.240 pairing of improved network connectivity and cloud computing. That removes 0:41:53.239 --> 0:41:56.319 the need for the device that you're interacting with to 0:41:56.400 --> 0:41:59.440 do all the processing on its own. So, if you 0:41:59.440 --> 0:42:01.799 think about or the history of computing, we used to 0:42:01.880 --> 0:42:05.160 do main frames with dumb terminals that attached the main frame, 0:42:05.440 --> 0:42:08.120 so the terminal wasn't doing any computing. It was just 0:42:08.200 --> 0:42:11.520 tapping into the mainframe computer, which was sending results back 0:42:11.520 --> 0:42:13.880 to the terminal. Then you get to the era of 0:42:13.960 --> 0:42:17.640 personal computers, where you had a device sitting on your 0:42:17.680 --> 0:42:20.560 desk that did all the computing and it didn't connect 0:42:20.560 --> 0:42:23.920 to anything else. Then we get up to networking and 0:42:23.960 --> 0:42:27.640 the Internet, where we suddenly had the capability of having 0:42:27.840 --> 0:42:31.720 really powerful computers or grids of computers that were able 0:42:31.760 --> 0:42:35.200 to take on processing power. Uh, and you just you 0:42:35.239 --> 0:42:38.719 send the request out to the Internet and you get 0:42:38.719 --> 0:42:42.080 the response back. That's the basis of cloud computing. So 0:42:43.200 --> 0:42:47.000 your your command or message or whatever relays back to 0:42:47.160 --> 0:42:50.759 servers on the cloud that then process it and send 0:42:50.760 --> 0:42:54.759 the proper response to whatever device you're interacting with, and 0:42:54.800 --> 0:42:57.120 then you get the result. So with the case of 0:42:57.120 --> 0:42:59.840 the smart speaker, it might be playing a specific so 0:43:00.000 --> 0:43:02.360 long or giving you a weather report or whatever it 0:43:02.440 --> 0:43:05.279 might be. Now, if the speakers were doing some of 0:43:05.320 --> 0:43:09.759 that computation themselves, that would be an example of edge computing, 0:43:10.160 --> 0:43:13.440 where the processing takes place at least in part, at 0:43:13.480 --> 0:43:16.799 the edge of a network at those end points. But 0:43:16.960 --> 0:43:20.240 for now, most of the implementations we see send data 0:43:20.280 --> 0:43:22.120 back to the cloud to get the right response, so 0:43:22.160 --> 0:43:25.240 you have to have a persistent Internet connection. These devices 0:43:25.280 --> 0:43:28.480 are not useful without that connection. You do have some 0:43:28.600 --> 0:43:32.360 smart speakers that can connect to another device like a 0:43:32.360 --> 0:43:36.320 smartphone via Bluetooth, so you could do things that way, 0:43:36.760 --> 0:43:40.920 but without those connections, the smart speaker turns into, you know, 0:43:41.040 --> 0:43:44.320 just a dumb speaker, or sometimes just a paperweight. Now, 0:43:44.840 --> 0:43:48.040 this collection of technologies and disciplines are what enabled Apple 0:43:48.440 --> 0:43:52.360 to introduce Sirie in two thousand and eleven, and Syria 0:43:52.440 --> 0:43:56.160 is a virtual assistant. Series origins actually trace back to 0:43:56.520 --> 0:44:00.480 the Stanford Research Institute and a group of guys Grouber, 0:44:00.560 --> 0:44:04.279 Adamshire and dog kit Louse who had been working on 0:44:04.320 --> 0:44:08.240 the concept since the nineteen nineties, and when Apple launched 0:44:08.239 --> 0:44:11.000 the iPhone in two thousand seven, they saw the iPhone 0:44:11.040 --> 0:44:14.680 as a potential platform for this virtual assistant that they 0:44:14.719 --> 0:44:17.520 had been building, and they thought, well, this is perfect 0:44:17.560 --> 0:44:20.879 because the iPhone has a microphone, so the assistant can 0:44:20.920 --> 0:44:23.719 respond to voice commands as a speaker, so it could 0:44:23.719 --> 0:44:26.200 communicate back to the user, it could do all sorts 0:44:26.239 --> 0:44:30.480 of stuff. We can tap into the interoperability of apps 0:44:30.520 --> 0:44:33.160 on the device. It's a perfect platform for us to 0:44:33.239 --> 0:44:36.839 deploy this. So they developed an app once the opportunity 0:44:36.880 --> 0:44:41.279 arose because apps were not available for development immediately when 0:44:41.320 --> 0:44:45.840 Apple launched the iPhone, and once they did launch that app, 0:44:46.719 --> 0:44:50.240 uh within a month, less than a month, Steve Jobs 0:44:50.280 --> 0:44:52.040 was on the phone calling them up and offering to 0:44:52.120 --> 0:44:54.560 buy the technology, which of course they would agree to 0:44:54.880 --> 0:44:57.920 and it would become an integrated component in Apple's iPhone 0:44:57.960 --> 0:45:02.480 line afterward. And that's where voice assistants kind of lived 0:45:02.640 --> 0:45:05.560 for a few years. They mostly lived on smartphones like 0:45:05.640 --> 0:45:10.080 the iPhone. But in November two thousand fourteen, Amazon introduced 0:45:10.160 --> 0:45:14.400 the Amazon Echo smart speaker, which was originally only available 0:45:14.440 --> 0:45:17.600 for Prime members, and it had its own virtual assistant 0:45:17.760 --> 0:45:22.879 named Alexa, and thus the smart speaker era officially began. Now, 0:45:22.920 --> 0:45:25.480 there are plenty of other smart speakers that are on 0:45:25.520 --> 0:45:28.600 the market these days. There are products from Google like 0:45:28.719 --> 0:45:31.879 Google Home. Uh, there are so no speakers that can 0:45:31.920 --> 0:45:35.920 connect to services like Amazon's Alexa or Google's Assistant, and 0:45:36.000 --> 0:45:38.799 we're probably going to see a ton more, both from 0:45:38.880 --> 0:45:42.120 companies that piggyback onto services from the big providers like 0:45:42.160 --> 0:45:45.160 Google and Amazon, and maybe some that are trying to 0:45:45.200 --> 0:45:47.360 make a go of it with their own branded virtual 0:45:47.400 --> 0:45:52.040 assistants and services. Smart speakers respond to commands after they 0:45:52.120 --> 0:45:55.880 quote unquote here a wake up word or phrase. Now, 0:45:56.120 --> 0:45:59.040 I'm gonna make up a wake up phrase right now 0:45:59.239 --> 0:46:02.440 so that I don't set off anyone's smart speaker or 0:46:02.480 --> 0:46:05.520 smart watch or smartphone or smart car or whatever it 0:46:05.600 --> 0:46:08.520 might be. So this is just a fictional example of 0:46:08.560 --> 0:46:11.719 a wake up phrase. So let's say I have a 0:46:11.719 --> 0:46:15.040 smart speaker and the wake up phrase for my smart 0:46:15.080 --> 0:46:18.680 speaker happens to be hey, they're Genie. Well, my smart 0:46:18.719 --> 0:46:21.120 speaker has a microphone, so it can detect when I 0:46:21.160 --> 0:46:27.480 say that, but really it's constantly detecting all sounds in 0:46:27.680 --> 0:46:31.000 its environment. The microphone is always active. It has to 0:46:31.040 --> 0:46:33.239 be in order to be able to pick up on 0:46:33.360 --> 0:46:38.160 when I say the wake up phrase. So the microphone 0:46:38.200 --> 0:46:41.480 is always active on most smart speakers. There's somewhere you 0:46:41.520 --> 0:46:44.160 can program it so that it will only activate if 0:46:44.160 --> 0:46:47.480 you first touch the speaker and that wakes it up. 0:46:47.840 --> 0:46:49.680 There's some that you can do that with, But for 0:46:49.719 --> 0:46:53.359 the most part, they're always listening. While the speaker can 0:46:53.480 --> 0:46:57.960 quote unquote here everything, it's not listening to everything. In 0:46:57.960 --> 0:47:01.200 other words, it's not mon of during the specific things 0:47:01.200 --> 0:47:03.640 being said. At least that's what we've been told. And honestly, 0:47:04.200 --> 0:47:07.320 that makes a ton of sense from an operational standpoint. 0:47:07.520 --> 0:47:10.080 And the reason I say that is that the sheer 0:47:10.120 --> 0:47:13.359 amount of information that would be flooding in from all 0:47:13.440 --> 0:47:16.919 the microphones on all the smart devices from any one 0:47:17.080 --> 0:47:20.160 provider that happened to be deployed all over the world, 0:47:20.360 --> 0:47:23.880 that would be an astounding amount of data. And sifting 0:47:23.920 --> 0:47:27.000 through all that data to find stuff that's useful would 0:47:27.040 --> 0:47:29.840 take an enormous amount of effort and time and and 0:47:29.960 --> 0:47:33.719 processing power. So while you could have all the microphones 0:47:33.760 --> 0:47:37.120 listening in all over the place, finding out who to 0:47:37.200 --> 0:47:39.680 listen to at what time would be a lot trickier 0:47:39.719 --> 0:47:41.840 and probably not worth the effort it would take to 0:47:41.880 --> 0:47:46.440 pull something like that off. So what these speakers and 0:47:46.560 --> 0:47:50.240 other devices are actually doing is looking for a signal 0:47:50.480 --> 0:47:53.640 that matches the one that represents the wake phrase. So 0:47:53.719 --> 0:47:57.799 when I say, hey, they're Genie, the microphone picks up 0:47:57.800 --> 0:48:00.920 my voice, which the mic then try inslates into an 0:48:00.920 --> 0:48:04.640 electrical signal which gets digitized and compared against the digital 0:48:04.640 --> 0:48:09.000 fingerprint of the predesignated wake up phrase. And in this case, 0:48:09.480 --> 0:48:13.880 the two phrases match. It's like a fingerprint matching something 0:48:13.920 --> 0:48:16.719 that was left at a site. So that turns the 0:48:16.760 --> 0:48:20.080 speaker into an active listener rather than a passive one. 0:48:20.120 --> 0:48:23.600 It's ready to accept a command or a question and 0:48:23.680 --> 0:48:27.200 to respond to me. But if I didn't say, hey, 0:48:27.280 --> 0:48:30.920 they're Genie, then the speaker would remain in passive mode 0:48:31.360 --> 0:48:35.239 because it wouldn't have a digital fingerprint that matches the 0:48:35.320 --> 0:48:38.400 one of the wake up phrase. Everything stays at the 0:48:38.440 --> 0:48:41.840 local level, and none of my sweet secret speech gets 0:48:41.880 --> 0:48:45.080 transmitt related across the internet. It's all staying right there. 0:48:45.560 --> 0:48:47.960 At least that's what we've been told. And again I 0:48:48.000 --> 0:48:50.719 don't have any reason to disbelieve this, but it is 0:48:50.760 --> 0:48:53.800 something to keep in mind. You are talking about devices 0:48:53.840 --> 0:48:55.960 that have microphones. Of course, if you have a smartphone, 0:48:56.000 --> 0:48:57.719 you've already got one of those or a cell phone. 0:48:57.719 --> 0:49:00.600 In general, you've got a device with a microphone on 0:49:00.680 --> 0:49:04.040 it neck near you pretty much all the time. Now, 0:49:04.680 --> 0:49:07.360 once I do make a request with my smart speaker, 0:49:07.560 --> 0:49:09.880 the speaker then sends that request up to the cloud 0:49:10.000 --> 0:49:14.120 where it gets processed, It's analyzed, uh, and then a 0:49:14.160 --> 0:49:18.080 proper response is returned to me, whether that is playing 0:49:18.080 --> 0:49:20.480 a song or giving me information I've asked for, or 0:49:20.520 --> 0:49:23.160 maybe even interacting with some other smart device in my home, 0:49:23.280 --> 0:49:26.880 such as adjusting the brightness of my smart lights in 0:49:26.960 --> 0:49:30.160 my house. Now, if the system is not sure about 0:49:30.200 --> 0:49:34.240 whatever it was I just said, it will probably return 0:49:34.400 --> 0:49:37.520 an error phrase. So maybe maybe I'm too far away 0:49:37.600 --> 0:49:40.759 from the speaker, so it's it couldn't quote unquote hear 0:49:40.840 --> 0:49:43.719 me really well. Or maybe I've got a mouthful of 0:49:43.760 --> 0:49:46.600 peanut butter or something as I want to do. Then 0:49:46.640 --> 0:49:48.600 I'm going to get something like I'm sorry, I don't 0:49:48.640 --> 0:49:50.480 know how to do that, or I'm sorry I didn't 0:49:50.520 --> 0:49:53.359 understand you, and then I'd have to repeat it. Now, 0:49:53.400 --> 0:49:57.040 smart speakers are pretty cool. However, they do represent another 0:49:57.080 --> 0:50:02.040 piece of technology that you have to network to other devices, 0:50:02.200 --> 0:50:06.120 including your own home network, and as such that means 0:50:06.200 --> 0:50:10.600 that they represent a potential vulnerability in a network. It 0:50:10.640 --> 0:50:14.520 doesn't mean they're automatically vulnerable, but it means that every 0:50:14.520 --> 0:50:18.319 time you are connecting something to your network, then you're 0:50:18.400 --> 0:50:23.960 creating another potential attack vector for a hacker. Right now, 0:50:24.000 --> 0:50:28.879 if everything is super strong, it it doesn't really effectively 0:50:29.040 --> 0:50:32.799 change your safety in any meaningful way. But if one 0:50:32.840 --> 0:50:35.880 of those things that you connect to your network is 0:50:36.000 --> 0:50:38.640 less strong than the others, you're looking at the weakest 0:50:38.680 --> 0:50:41.520 link situation where a hacker with the right know how 0:50:41.560 --> 0:50:45.280 in tools could potentially target that part of your network 0:50:45.600 --> 0:50:49.320 to get entry into everything else. And when you're talking 0:50:49.360 --> 0:50:53.120 about a smart speaker, you're talking about device that has 0:50:53.400 --> 0:50:57.440 an active microphone on it. So potentially, if someone were 0:50:57.480 --> 0:50:59.879 able to compromise a smart speaker, they would be able 0:50:59.920 --> 0:51:03.120 to listening on anything that was within range of that 0:51:03.200 --> 0:51:07.920 smart speakers microphone. So that's why you have to at 0:51:07.960 --> 0:51:11.759 least be cognizant of that, do your research, make sure 0:51:11.800 --> 0:51:15.000 the devices you're connecting to your network are rated well 0:51:15.040 --> 0:51:18.640 as from a security standpoint, when you're setting things up 0:51:18.680 --> 0:51:22.239 and you have to create passwords, create strong passwords that 0:51:22.320 --> 0:51:26.200 are not used anywhere else. The harder you make things 0:51:26.440 --> 0:51:30.160 the more likely hackers will just pass you by, not 0:51:30.239 --> 0:51:33.919 because you're too tough to crack. Never get your into 0:51:33.960 --> 0:51:37.200 your head that you're too strong to to be hacked, 0:51:37.440 --> 0:51:41.160 but rather if there's someone who's weaker than the hackers 0:51:41.160 --> 0:51:43.600 are going to go after that person instead. So just 0:51:43.680 --> 0:51:48.640 don't be the weak person. Practice really good security behaviors, 0:51:48.840 --> 0:51:53.279 and you're more likely to discourage attackers and they'll they'll 0:51:53.320 --> 0:51:57.359 go on to someone else. Um, especially if you're talking 0:51:57.360 --> 0:51:59.960 about newbies who don't really know their way around their 0:52:00.080 --> 0:52:02.919 just using tools that other people have designed. They get 0:52:02.920 --> 0:52:05.319 discouraged very quickly. They'll move on to someone else because 0:52:05.360 --> 0:52:09.359 there's always another potential target. I'm curious about you guys, 0:52:09.360 --> 0:52:12.720 whether or not you have any smart speakers in your life, 0:52:13.200 --> 0:52:15.919 and uh if you find them useful. I find mine 0:52:15.960 --> 0:52:20.160 pretty useful. I use it for a very narrow range 0:52:20.320 --> 0:52:23.480 of things. I don't tend to use it. I definitely 0:52:23.520 --> 0:52:25.680 don't use it to its full potential. I know that 0:52:26.239 --> 0:52:29.080 because what's in the blue moon. I'll just try something 0:52:29.120 --> 0:52:31.719 and I'm amazed at what happens when when I get 0:52:31.719 --> 0:52:34.759 a response. But for the most part, I'm asking about 0:52:34.800 --> 0:52:38.080 whether what I can feed my dog whether or not 0:52:38.120 --> 0:52:41.040 it can turn on the lights and uh and and 0:52:41.760 --> 0:52:45.319 that's about it. Are occasionally playing a song. Um, but 0:52:45.480 --> 0:52:47.759 I'm curious what you guys are using them for. Reach 0:52:47.800 --> 0:52:50.480 Out to me on social networks on Facebook and I'm 0:52:50.520 --> 0:52:52.480 on Twitter, and the handle for both of those is 0:52:52.520 --> 0:52:56.880 text stuff. H s W also use that those handles 0:52:56.880 --> 0:52:59.320 if you have suggestions for future episodes. If you've got, 0:52:59.440 --> 0:53:02.120 you know, an idea for either a company or a 0:53:02.160 --> 0:53:05.640 technology or a theme in tech you'd really like me 0:53:05.719 --> 0:53:08.759 to tackle, let me know there and I'll talk to 0:53:08.760 --> 0:53:16.279 you again really soon. Text Stuff is a production of 0:53:16.280 --> 0:53:19.319 I Heart Radio's How Stuff Works. For more podcasts from 0:53:19.360 --> 0:53:23.120 my heart Radio, visit the i heart Radio app, Apple Podcasts, 0:53:23.280 --> 0:53:25.240 or wherever you listen to your favorite shows.