Creepy Speaking Neural Networks

March 22, 2017 by 37 Comments

Tech artist [Alexander Reben] has shared some work in progress with us. It’s a neural network trained on various famous peoples’ speech (YouTube, embedded below). [Alexander]’s artistic goal is to capture the “soul” of a person’s voice, in much the same way as death masks of centuries past. Of course, listening to [Alexander]’s Rob Boss is no substitute for actually watching an old Bob Ross tape — indeed it never even manages to say “happy little trees” — but it is certainly recognizable as the man himself, and now we can generate an infinite amount of his patter.

Behind the scenes, he’s using WaveNet to train the networks. Basically, the algorithm splits up an audio stream into chunks and tries to predict the next chunk based on the previous state. Some pre-editing of the training audio data was necessary — removing the laughter and applause from the Colbert track for instance — but it was basically just plugged right in.

The network seems to over-emphasize sibilants; we’ve never heard Barack Obama hiss quite like that in real life. Feeding noise into machines that are set up as pattern-recognizers tends to push them to the limits. But in keeping with the name of this series of projects, the “unreasonable humanity of algorithms”, it does pretty well.

He’s also done the same thing with multiple speakers (also YouTube), in this case 110 people with different genders and accents. The variation across people leads to a smoother, more human sound, but it’s also not clearly anyone in particular. It’s meant to be continuously running out of a speaker inside a sculpture’s mouth. We’re a bit creeped out, in a good way.

We’ve covered some of [Alexander]’s work before, from the wince-inducing “Robot Bites Man” to the intellectual-conceptual “All Prior Art“. Keep it coming, [Alexander]!

Continue reading “Creepy Speaking Neural Networks” →

Making More Of Me Money

March 22, 2017 by 28 Comments

For the last few years, Hackaday has really been stepping up our game with marketing materials. Our t-shirts and swag are second to none, and last year we introduced the ‘Benchoff Buck’ (featured above), a bill replete with Jolly Wrencher EURions that is not yet legal currency. At least until we get a sweet compound in the desert, that is.

[Andrew Sowa] created the Benchoff Nickel. It’s a visage of yours truly emblazoned on a PCB, rendered in FR4, silkscreen, gold, and OSHPark’s royal purple. In doing so, [Andrew] has earned himself a field commission to the rank of lieutenant and can now reserve the dune buggy for a whole weekend.

The Benchoff Nickel was created in KiCad using the Bitmap2Component functionality. Planning this required a little bit of work; there are only five colors you can get on an OSH Park PCB, from white to gold to beige to purple (soldermask on top of copper) to black (soldermask with no copper). Luckily, the best picture we have of me renders very well in five colors.

The Bitmap2Component part of KiCad will only get you so far, though. It’s used mainly to put silkscreen logos on a board, and messing around with copper and mask layers is beyond its functionality. To import different layers of my face into different layers of a KiCad PCB, [Andrew] had to open up Notepad and make a few manual edits. It’s annoying, but yes, it can be done.

OSH Park’s fabs apparently use two different tones of FR4

The Benchoff Nickel can be found on Github and as a shared project on OSH Park ($22.55 for three copies). One little curiosity of the OSH Park fabrication process presented itself with [Andrew]’s second order of Benchoff Nickels. OSH Park uses at least two board houses to produce their PCBs, and one of them apparently uses a lighter shade of FR4. This resulted in a lighter skin tone for the second order of Benchoff Nickels.

This is truly tremendous work. I’ve never seen anything like this, and it’s one of the best ‘artistic’ PCBs I’ve ever held in my hands. It was a really great surprise when [Andrew] handed me one of these at the Hackaday Unconference in Chicago. I’ll be talking to [Andrew] again this week at the Midwest RepRap festival, and we’re going to try and figure out some way to do a small run of Benchoff Nickels.

Edit: OSH Park revealed why there are different tones of FR4. In short, there aren’t. The lighter shade of skintone is actually FR408, which is used on 4-layer boards.

Saturday clock - 1 CPU clock cycle per day

Saturday Clock: An 0.000011574Hz ATtiny85 Clock

March 21, 2017 by 40 Comments

In these times when we try to squeeze out extra clock cycles by adding more cores to our CPUs and by enlisting the aid of GPUs, [Ido Gendel] thought it would be fun to go in the exact opposite direction, supply a clock to the ATtiny85 that cycles only once per day, or at 0.000011574Hz. What application could this have? Well, if he could do it in seven instructions or less, how about turning on an LED at sunset Friday evening, to indicate the start of the Jewish Shabbat (Saturday), and turn it off again at sunset Saturday evening.

Notice the subtlety. A clock that cycles once per day means you can execute at most one instruction per day. Luckily on AVR microcontrollers, the instructions he needed can execute in just one cycle. That of course meant diving down into assembly code. [Ido] wasn’t an assembly wizard, so to find the instructions, he compiled C code and examined the resulting assembly until he found what he needed. One instruction turns on the LED and the instruction immediately following turns it off again, which normally would make it happen too fast for the human eye to register. But the instruction to turn it on runs on Friday evening and the very next instruction, the one that turns it off, doesn’t run until Saturday evening. Do you feel like you’re in a science fiction story watching time slowed down? Freaky. A few NOPs and the jump for the loop take up the remaining five cycles for the week.

For the source of the clock he chose to use an LDR to detect when the light level dropped at the end of the day. The problem he immediately ran into was that clouds, bird shadows, and so on, also cause drops in the light level. The solution he found was to widen the light and dark range by adding a TLV3702 push-pull output comparator and some resistors. [Ido] gives a detailed explanation of the circuit in the video after the break.

Continue reading “Saturday Clock: An 0.000011574Hz ATtiny85 Clock” →

Tiny Electric Motor Runs On Power From An LED

March 21, 2017 by 31 Comments

If you were not aware, LEDs can also work in reverse: they deliver tiny amounts of current, in the microamp range, when illuminated. If you look on YouTube you can find several videos of solar panels built with arrays of LEDs, but powering an electric motor with a single 3 mm LED is something that we’ve never seen before. [Slider2732] built a small electric motor that happily runs from a green LED in sunlight.

The motor uses four coils of 1,000 ohms each. Using coils with many turns of very fine wire helps to draw less current while keeping an appropriate magnetic field for the motor to run. To keep friction at a minimum, the rotor uses a needle that hangs from a magnet. Four neodymium magnets around the rotor are periodically pushed by the coils, generating rotation. A simple two-transistor circuit takes care of the synchronization and yes, the motor does run on the four microamps provided by the LED, and runs pretty well.

Building motors is definitely an enjoyable activity, these small pulse motors can be built in just a couple of hours. You can use coils with just a few tens of turns which are much more easy to make but of course you will need something more than four microamps! The nice part of making an ultralow current motor like this is that it can run for a very long time on a tiny battery or even a capacitor, we invite you to try building one.

Continue reading “Tiny Electric Motor Runs On Power From An LED” →

Bomb Defusal Fun With Friends!

March 21, 2017 by 16 Comments

Being a member of the bomb squad would be pretty high up when it comes to ranking stressful occupations. It also makes for great fun with friends. Keep Talking and Nobody Explodes is a two-player video game where one player attempts to defuse a bomb based on instructions from someone on the other end of a phone. [hephaisto] found the game great fun, but thought it could really benefit from some actual hardware. They set about building a real-life bomb defusal game named BUMM.

The “bomb” itself consists of a Raspberry Pi brain that communicates with a series of modules over a serial bus. The modules consist of a timer, a serial number display, and two “riddle” boxes covered in switches and LEDs. It’s the job of the bomb defuser to describe what they see on the various modules to the remote operator, who reads a manual and relays instructions based on this data back to the defuser. For example, the defuser may report seeing a yellow and green LED lit on the riddle module – the operator will then look this up and instruct the defuser on which switches to set in order to defuse that part of the bomb. It’s the challenge of quickly and accurately communicating in the face of a ticking clock that makes the game fun.

[hephaisto] took this build to Make Rhein-Main 2017, where they were very accepting of a “bomb” being brought onto the premises. The game was setup in a booth with an old analog S-video camera feed and a field telephone for communication – we love the detail touches that really add atmosphere to the gameplay experience.

Overall, it’s a great project that could easily be recreated by any hackerspace looking for something fun to share on community nights. The build files are all available on the project GitHub so it’s easy to see the nuts and bolts of how it works.

We’ve seen builds that bring video games into the real world before – like this coilgun Scorched Earth build. Fantastic.

Your VR Doesn’t Stink (Yet)

March 21, 2017 by 24 Comments

What does it smell like when the wheels heat up on that Formula 1 car you drive at night and on the weekends? You have no idea because the Virtual Reality experience that lets you do so doesn’t come with a nasal component. Yet.

Shown here is an olfactory device that works with Oculus Rift and other head-mounted displays. The proof of concept is hte work of [Kazuki Hashimoto], [Yosuke Maruno], and [Takamichi Nakamoto] and was shown of at last year’s IEEE VR conference. It lets the wearer smell the oranges when approaching a tree in a virtual environment. In other words, it makes your immersive experience smelly.

As it stands this a pretty cool little device which atomizes odor droplets while a tiny fan wafts them to the wearer’s nose. There is a paper which presumably has more detail but it’s behind a pay wall so for now check out the brief demo video below. Traditionally an issue with scent systems is the substance stuck in the lines, which this prototype overcomes with direct application from the reservoir. Yet to be solved is the availability for numerous different scents.

This build came to our attention via an UploadVR article that does a good job of covering some of the scent-based experiments over the years. They see some of the same hurdles we do: odors linger and there is a limited palette that can be produced. We assume the massive revenue of the gaming industry is going to drive research in this field, but we won’t be lining up to smell gunpowder and dead bodies (or rotting zombies) anytime soon.

The more noble effort is in VR applications like taking the elderly and immobile back for another tour of places they’ll never again be able to visit in their lives. Adding the sense of smell, which has the power to unlock so many memories, makes that use case so much more powerful. We think that’s something everyone can be hopeful about!

Continue reading “Your VR Doesn’t Stink (Yet)” →

Negative Resistance: It Shouldn’t Make Sense!

March 21, 2017 by 55 Comments

When you leaf through a basic electronics textbook, you’ll find chapters describing in detail the operation of the various components. Resistors, capacitors, inductors, and semiconductors. The latter chapter will talk about P and N type regions, introduce us to the diode, and then deal with the transistor: its basic operation, how to bias it, and the like.

A tunnel diode amplifier circuit. Chetvorno [CC0]
Particularly if your textbook is a little older, you may find a short section talking about the tunnel diode. There will be an odd-looking circuit that seems to make no sense at all, an amplifier formed from just a forward-biased diode and a couple of resistors. This logic-defying circuit you are told works due to the tunnel diode being of a class of devices having a negative resistance, though in the absence of readily available devices for experimentation it can be difficult to wrap your head around.

We’re all used to conventional resistors, devices that follow Ohm’s Law. When you apply a voltage to a resistor, a current flows through it, and when the voltage is increased, so does the current. Thus if you use a positive resistance device, say a normal resistor, in both the top and the bottom halves of a potential divider, varying the voltage fed into the top of the divider results in the resistor behaving as you’d expect, and the voltage across it increases.

In a negative resistance device the opposite is the case: increasing the voltage across it results in decreasing current flowing through it. When a large enough negative resistance device is used in the lower half of a resistive divider, it reduces the overall current flowing through the divider when the input voltage increases. With less current flowing across the top resistor, more voltage is present at the output. This makes the negative resistor divider into an amplifier.

The tunnel diodes we mentioned above are probably the best known devices that exhibit negative resistance, and there was a time in the early 1960s before transistors gained extra performance that they seemed to represent the future in electronics. But they aren’t the only devices with a negative resistance curve, indeed aside from other semiconductors such as Gunn diodes you can find negative resistance in some surprising places. Electrical arcs, for example, or fluorescent lighting tubes.

A typical negative resistance I-V curve. Chetvorno [CC0]
The negative resistance property of electric arcs in particular produced a fascinating device from the early twentieth century. The first radio transmitters used an electric arc to generate their RF, but were extremely inefficient and wideband, causing interference. A refinement treated the spark not as the source of the RF but as the negative resistance element alongside a tuned circuit in an oscillator, These devices could generate single frequencies at extremely high power, and thus became popular as high-powered transmitters alongside those using high-frequency alternators until the advent of higher powered tube-based transmitters around the First World War.

It’s unlikely that you will encounter a tunnel diode or other similar electronic component outside the realm of very specialist surplus parts suppliers. We’ve featured them only rarely, and then they are usually surplus devices from the 1960s. But understanding something of how they operate in a circuit should be part of the general knowledge of anyone with an interest in electronics, and is thus worth taking a moment to look at.

1N3716 tunnel diode header image: Caliston [Public domain].