split stenograph with numpad

[hagna]

This is for a split (two keyboards) stenograph style keyboard with a numpad on the right hand. Steno doesn't offer an improvement for numeric input, so I'm adding a numpad for the right hand.

Here is the steno layout. For more information on stenography see http://openstenoproject.org/

ploverdemokb.png (4.81 KiB) Viewed 10038 times

Here is the proposed new layout with numpad on the right hand: (what program do you use to post keyboard layouts inline?)
left hand: http://www.keyboard-layout-editor.com/# ... 232f6f9d6f
right hand: http://www.keyboard-layout-editor.com/# ... 6c5171746c

This plate ought to work for both left and right.

1st take

grid.jpg (55.91 KiB) Viewed 10043 times

It should make a fine stenograph for the half dozen people in the world who want that, but I want to use it as an easier to learn chording keyboard similar to the BAT (http://www.infogrip.com/bat-keyboard.html)

What do you think about this idea?

The first prototype might prompt further comments, so I'll post it later this week.

[Zekromtor]

Am I missing something here? A stenograph uses chords to type full words, that BAT keyboard you linked uses key combos to type single letters, which in no way shape or form is more efficient than normal typing like they claim.

Which are you trying to replicate?

[arbitrarycombination]

Nice idea. Looks like a more practical variant of a steno keyboard.
The lower row of the right hand is shared with the upper numpad row, right? Do you have to toggle numeric input or do use it with the asterisk key?

What are you going to do with the mounting holes that are left empty?

[hagna]

A stenograph uses chords to type full words, that BAT keyboard you linked uses key combos to type single letters, which in no way shape or form is more efficient than normal typing like they claim.

Which are you trying to replicate?

Somewhere in between. The plan is to make words out of phonemes using a dictionary for spelling like steno does. This makes it easier to learn than steno and more efficient than BAT. So the BAT claim of 50 WPM is false?

[hagna]

Nice idea. Looks like a more practical variant of a steno keyboard.
The lower row of the right hand is shared with the upper numpad row, right? Do you have to toggle numeric input or do use it with the asterisk key?

It's a toss up between that or reading the operators mind with an ECG.

What are you going to do with the mounting holes that are left empty?

Put legs in them? What do you mean?

[Zekromtor]

A stenograph uses chords to type full words, that BAT keyboard you linked uses key combos to type single letters, which in no way shape or form is more efficient than normal typing like they claim.

Which are you trying to replicate?

Somewhere in between. The plan is to make words out of phonemes using a dictionary for spelling like steno does. This makes it easier to learn than steno and more efficient than BAT. So the BAT claim of 50 WPM is false?

50 wpm max sounds about right, which is about 1/6th the max speed of a real steno and 1/3rd the max speed of a normal keyboard.

[trishume]

If you are looking for something easy to learn I suggest you check out the Velotype (http://velotype.com/en/) system instead. It doesn't allow you to type quite as fast as steno but it is much easier to learn and doesn't require a dictionary.

Also chording keyboards require lighter key activation force to be comfortable. You can lower cherry MX Red switch force to about 25g by cutting the 1.5cm springs down to 1cm. This gives keys which don't feel nice for normal typing but feel much more comfortable when typing multiple keys at once.

I'm currently building a Velotype based chording keyboard and have just finished the case. I just have to write the firmware and wire up the matrix and controller. It has way more keys than are necessary so that it can also be used with normal layouts, only the main chording keys have the modified switches. I've attached a picture, the CAD files can be found at https://github.com/trishume/KeyboardCAD.

[jacobolus]

That Velotype page says “Very light touch! Only 15 grams per key.” — what kind of switches are they using? €1500 is a bit too steep for me to buy one to find out.

[trishume]

That Velotype page says “Very light touch! Only 15 grams per key.” — what kind of switches are they using? €1500 is a bit too steep for me to buy one to find out.

I've talked to the guys who make it a bit. They use silicone dome switches with a custom sheet and PCB. The steep price is part of the reason for building my own, mine only cost about $250.

I'm also currently writing a bunch of computer language models and typing simulations to see if I can come up with a better Velotype-like chording system. The theoretical efficiency of orthographic syllable typing is 3.2 characters per stroke but the Velotype layout only achieves 2.7 without all sorts of extra rules. Stenotype achieves 3.8, but the system is more complex and it includes use of memorized short forms of long words.

[jacobolus]

If you can figure out a good model for typing English which is both (a) efficient and (b) reasonably intuitive (and ideally based on orthography rather than phonetics), I’d love to hear about it. I think the best bet is to aim for 3 keys each with middle and ring finger, 5-6 keys with each index finger, 4-5 with each pinky, and 4-5 with each thumb (and maybe some palm keys, though I'm not totally convinced about those). I think your current pictured keyboard has too many keys to comfortably reach for routine typing, though the extras might be fine for special occasional functions.

[hagna]

Also chording keyboards require lighter key activation force to be comfortable. You can lower cherry MX Red switch force to about 25g by cutting the 1.5cm springs down to 1cm. This gives keys which don't feel nice for normal typing but feel much more comfortable when typing multiple keys at once.

7bit just came through on the cherry reds! I'm not up to cutting ~40 springs though. The scissor switches in the lenovo thinkpad would work just great, but then I wouldn't be welcome in the mechanical keyboard club.

I'm currently building a Velotype based chording keyboard and have just finished the case. I just have to write the firmware and wire up the matrix and controller. It has way more keys than are necessary so that it can also be used with normal layouts, only the main chording keys have the modified switches. I've attached a picture, the CAD files can be found at https://github.com/trishume/KeyboardCAD.

That's a great looking case!

[trishume]

@jacobolus only 3 keys per finger (5 for index) are actually used for chording, the rest are just extra since my switch bag had 110 switches and so that I could also use it as a normal keyboard.

I also just did some simulations and came up with a simple autocomplete-based chording model that gets on average 4.26 characters per stroke / 1.18 strokes per word based on the Google n-gram corpus. It requires incorporating visual feedback from the screen though so it probably can't be used for typing at 200WPM but it has great economy of motion for RSI reduction (which is the main reason I made a chording keyboard). It's not quite finished though, the numbers are theoretical and I haven't actually turned it into a full layout yet.

@hagna modifying the switches doesn't actually take that long. It took me about 1:30 minute per switch with 2 toothpicks to open the switch, a ruler to measure the length, and a pair of wire snippers. Modifying all 46 only took about 1 hour and 30 minutes while listening to a podcast. The case is layered laser cut acrylic with 1/8" black acrylic on the top and bottom and 1/4" clear for the middle layers.

[jacobolus]

Hmm. Unfortunately I hate hate hate autocomplete.

Can you share details about your chord/layout design though?

[Zekromtor]

+1 to switch stuff not taking as long as you'd think. I opened, ultrasonic cleaned, dried, reassembled 90 alps switches recently and with some proper TV in the background it wasn't painful at all.

Great project, and great looking keyboard. Tell us more

[trishume]

Hmm. Unfortunately I hate hate hate autocomplete.

Can you share details about your chord/layout design though?

At the moment I only have a bunch of modelling programs that run on word frequency data. I have one that does a full finger allocation simulation of the Velotype layout, and a few that operate on a theoretically perfect (not actually simulated) layout where there are always enough fingers to stroke a syllable. My next step is to use some other data to develop a layout like Velotype but that generates less finger collisions where one finger would have to type two keys to stroke a syllable.

Also depending on why you hate autocomplete this system might not be bad. The way it works is that after every stroke in a word it uses the most frequent word that starts with those strokes. If the word shown is not the one you want then you simply keep stroking out the word like you would normally, otherwise you move on to the next word. If you want to type a short word like "fir" and it completes to "first" you just hit one palm key and it will end the stroke where you typed and not do any completion. It does require looping visual feedback into the typing process though.

I do have an eye tracker arriving soon so I might use gaze data to drive autocomplete so that you can type along with all muscle memory and no visual feedback. When it finds a completion the line of text will split into a tree and you just keep looking at the branch with the word you want and additional words will appear there and the others will disappear.

Here's the statistics running on the Google n-gram corpus:
Pure Velotype system without layout: 3.2 characters per stroke = 1.49 strokes per word
Velotype system with layout collisions: 2.98 characters per stroke = 1.622 strokes per word
Velotype with extra suffixes no layout (er|ed|en|e|ion|able|ing|al): 3.46 characters per stroke = 1.38 strokes per word
Stenotype with briefs and shortest dictionary entries: 3.74 characters per stroke = 1.14 strokes per word
Stenotype no briefs and longest dictionary entries: 3.0 characters per stroke = 1.60 strokes per word
Velotype with autocomplete based on the first strokes (no layout): 4.26 characters per stroke = 1.18 strokes per word

[hagna]

@trishume The time it takes to learn the new layout is very important too--at least it is for me. Have you seen any data on how long it takes to learn velotype? Steno is the worst for this:

"Most people in steno school get up to 100 WPM within the first six months or so. Then it takes them between two and six years to get the other 125 WPM"[0]

[0] http://opensource.com/life/11/12/open-s ... g-impaired

The model I like is based on A minimum chord stenograph keyboard for blind and sighted users by Hu, Zhongzhi. The layout presented in the paper takes comparably less time to learn than qwerty or steno (he makes the comparison in the paper). One of the subjects reached 59 WPM (295 characters/minute) in 60 hours of training. Also, the word "stop" would require 2 strokes.

I'd like to shoot for the following goals with the split stenograph with numpad:
1. In under 100 hours of practice people can type around 100 WPM without looking at the keyboard.
2. Able to use right handed at slower speed with no additional training.

[jacobolus]

Hmm. Unfortunately I hate hate hate autocomplete.

Also depending on why you hate autocomplete this system might not be bad.

I want my input to be explicit. I don’t want my computer guessing at what I intended to input, and I don’t want to have to think about whether the computer had its own interpretation.

All those automatic features which turn lines starting with asterisks into richly formatted bullet lists, or replace my periods with ellipsis characters, or which autocorrect what the machine assumes are spelling errors drive me up the wall.

I want the computer to provide me extra data and context, like the little squiggly red lines under misspelled words, or easy access to documentation about the functions I’m calling in a block of code. I don’t want the computer trying to outsmart me.

Here's the statistics running on the Google n-gram corpus:
Pure Velotype system without layout: 3.2 characters per stroke = 1.49 strokes per word
Velotype system with layout collisions: 2.98 characters per stroke = 1.622 strokes per word
Velotype with extra suffixes no layout (er|ed|en|e|ion|able|ing|al): 3.46 characters per stroke = 1.38 strokes per word
Stenotype with briefs and shortest dictionary entries: 3.74 characters per stroke = 1.14 strokes per word
Stenotype no briefs and longest dictionary entries: 3.0 characters per stroke = 1.60 strokes per word
Velotype with autocomplete based on the first strokes (no layout): 4.26 characters per stroke = 1.18 strokes per word

To be honest, I think the velotype sounds fine here. A bunch of unpredictability isn’t worth that kind of improvement. (But then again, I’m not a professional transcriptionist.)

[trishume]

The model I like is based on A minimum chord stenograph keyboard for blind and sighted users by Hu, Zhongzhi. The layout presented in the paper takes comparably less time to learn than qwerty or steno (he makes the comparison in the paper). One of the subjects reached 59 WPM (295 characters/minute) in 60 hours of training. Also, the word "stop" would require 2 strokes.

Cool, I'm reading that paper now. If it looks promising/easy to implement I will write a model for it so that I can compare it with the other layouts. I don't have good info on how long it takes to learn Velotype but I just sent an email to one of the people who work there asking about that. I do know that they claim it is significantly easier to learn than stenotype and it certainly has fewer rules. I also seem to recall something about the Velotype course being 6 months long versus many years for stenotype, but I don't know what the graduation WPM is for Velotype.

Also the steno 6 months may or may not be misleading, http://stenoknight.com/StenoSchool.html says it took her a year before her first speed test at 50wpm. But her info might be different from the normal, or the dates might be wrong.

What is the main reason you want to learn a chorded layout hagna? For me it is mostly economy of motion for ergonomic reasons but speed is a bonus. Do you have a top speed goal? Different goals might mean different optimal layouts.

Velotype it is possible to type like a normal keyboard using only one hand. You can actually type single letters with either hand but I'm planning on getting rid of this redundancy to have single letters on one hand be punctuation marks for coding.

One cool thing about chording keyboard designs is that most of them have interchangeable physical designs. Both steno and velotype have 4 states per finger even though they have different amounts of keys, because steno allows pressing two keys with one finger. So even though I have my full keyboard hardware, I haven't decided on a layout yet. Also if I end up liking chording layouts I might upgrade to one of these http://infinitytraditional.com/Productd ... aspx?ID=15, it has hall effect adjustable force smart switches!

[hagna]

Plate mounted keyboards aren't too expensive, so I wonder how much it would cost to build one of these http://infinitytraditional.com/Productd ... aspx?ID=15. Yes, I agree maybe not a logical conclusion.

I don’t want the computer trying to outsmart me.

Yes I'm with you on that. That's why I like the dictionary approach/hack for translating keystrokes to output. You are in charge of your dictionary.

What is the main reason you want to learn a chorded layout hagna? For me it is mostly economy of motion for ergonomic reasons but speed is a bonus. Do you have a top speed goal? Different goals might mean different optimal layouts.

The main reason is that I can type fast without looking at my keys and sometimes win at typeracer, BUT only after years and years (at least 10 lets say) of keyboard experience. I'm using QWERTY. As with steno or shorthand a human can work hard to achieve great results i.e. 300 WPM 90 % accuracy, so why not apply that same idea just enough to get to 100 WPM. So the main reason is speed, and that is both speed to learn and speed to type. Another main reason that falls out of that paper is one handed mobile input. Another reason is input without looking at the keys. Ok 3 main reasons: speed to learn and to type, one handed, and using the keys without looking at them.

[trishume]

Plate mounted keyboards aren't too expensive, so I wonder how much it would cost to build one of these http://infinitytraditional.com/Productd ... aspx?ID=15. Yes, I agree maybe not a logical conclusion.

I think in order to make a nice lever based design the levers would have to be made out of metal, and custom metal fabrication is EXPENSIVE. So I don't think I would want to try and design a DIY lever based steno machine.

Yes I'm with you on that. That's why I like the dictionary approach/hack for translating keystrokes to output. You are in charge of your dictionary.

The problem is that you have to make a dictionary, and that's a time-consuming endeavour. Also you have to do all sorts of extra worrying about homophones like "won" vs. "one".

The main reason is that I can type fast without looking at my keys and sometimes win at typeracer, BUT only after years and years (at least 10 lets say) of keyboard experience. I'm using QWERTY. As with steno or shorthand a human can work hard to achieve great results i.e. 300 WPM 90 % accuracy, so why not apply that same idea just enough to get to 100 WPM. So the main reason is speed, and that is both speed to learn and speed to type. Another main reason that falls out of that paper is one handed mobile input. Another reason is input without looking at the keys. Ok 3 main reasons: speed to learn and to type, one handed, and using the keys without looking at them.

I finished reading that paper on the minimum chord stenograph. The thing is it has data for how long it takes to learn, but none on maximum speeds. The thing about most chorded layouts is that they are harder to learn than QWERTY but have a higher maximum speed, whereas this system looks like it would produce only small gains if any of maximum speed, the efficiency looks to be around 2.5 characters per stroke. If you type qwerty at 50-80WPM now then less than 100 hours of practice would get you up to the 100WPM you seek. It would be interesting to see what kind of speeds you could attain with the minimum chord system but there's little evidence they would be that much better than QWERTY.

If you want higher than that you will need a layout that takes advantage of the parallelization of muscle memory, at 100WPM QWERTY takes 8 ordered movements per second and Velotype takes 2, so you don't hit the upper bound of muscle movement coordination.

Another advantage of chorded layouts that you lose with minimum chord steno is since so many of the 4096 possible strokes produce valid combinations you can't as easily bind unused strokes to actions like "move window right" or phrase short-forms.

[jacobolus]

I think in order to make a nice lever based design the levers would have to be made out of metal, and custom metal fabrication is EXPENSIVE. So I don't think I would want to try and design a DIY lever based steno machine.

I think he might just mean that key layout... not necessarily built with levers.

The problem is that you have to make a dictionary, and that's a time-consuming endeavour. Also you have to do all sorts of extra worrying about homophones like "won" vs. "one".

Yeah, which is why I want an orthography based system rather than a phonetic one, at least for English. For Chinese or something a phonetic system might be better.

As for making the dictionary: it can be done bit by bit over years. For instance, the way this guy does: http://geekhack.org/index.php?topic=548 ... msg1235489
http://proword-transcription.blogspot.com.au

[Squeaky Wheel]

How hard would it be to mod an ergodox to make something velotype-ish?

[xman]

... Also if I end up liking chording layouts I might upgrade to one of these http://infinitytraditional.com/Productd ... aspx?ID=15, it has hall effect adjustable force smart switches!

Getting an error with that link.
I'd like to see what you mean, so please post another link.
Thanks.

[trishume]

Here's a fixed link http://www.infinitytraditional.com/infi ... c-student/, they updated their website since I last posted.

The thing about modding an ergodox is that it doesn't necessarily have the right arrangement. You might get by with just modifying the switches and then maybe adding some palm keys, but it is certainly better to just modify the cad files to make a custom keyboard.

In other news, I finished all the soldering for my keyboard, now I just have to write the firmware to support chorded typing.

[xman]

Hi hagna,
Interesting project. Thanks for posting.
Looking forward to seeing more.

[xman]

Here's a fixed link http://www.infinitytraditional.com/infi ... c-student/, they updated their website since I last posted.

The thing about modding an ergodox is that it doesn't necessarily have the right arrangement. You might get by with just modifying the switches and then maybe adding some palm keys, but it is certainly better to just modify the cad files to make a custom keyboard.

In other news, I finished all the soldering for my keyboard, now I just have to write the firmware to support chorded typing.

____________________________________________________________

Hi trishume,

Thanks for that link.
I like fully split keyboards for ergonomic reasons,
just my personal preference.

When you consider pressing 2 or more keys using the pinkys,
I don't think 20g is light enough, not for _my_ pinkys,
not when chording. (Maybe OK for the other fingers.)

If you cut the 2 non-compressible ends off the coiled spring 1st,
you can get to a lower actuation force. I find the shortest length
I can do is ~9mm or ~8.8mm which may result in around 15g.

Interesting project.
Thanks for posting.

[trishume]

I just wrote up a build log with more info about my keyboard: http://thume.ca/2014/09/08/creating-a-k ... -hardware/

@xman You should never have to press more than 2 keys with one finger on any decent chording layout, in fact on the Velotype layout theoretically you never have to press multiple keys with one finger.

Also I don't see how 20g could be to heavy because 2 keys * 20 grams is 40 grams which is the weight of a normal key. So unless you only type on low force keyboards you should be able to easily press 40 grams with your pinky finger.

Your non-compressible end cutting technique is interesting, I hadn't thought of that before. My current keys are plenty light for me (and already finished) but I might try that with some of my extra switches, just out of curiosity about what a SUPER light switch feels like.

You've obviously worked on low force switches before, what kind of things have you done with them?

[xman]

t> @xman You should never have to press more than 2 keys with one finger on any decent chording layout, in fact on the Velotype layout theoretically you never have to press multiple keys with one finger.


Hi trishume,
I didn't know about the Velotype 1 key per finger layout.
I thought Velotype was closer to steno, but I was obviously wrong.
It must be a lot different than the keyboard stenography layout
(where most fingers are sometimes often pressing 2 keys each,
except the right pinky sometimes has to press 4 keys at once).


t> Your non-compressible end cutting technique is interesting, I hadn't thought of that before. My current keys are plenty light for me (and already finished) but I might try that with some of my extra switches, just out of curiosity about what a SUPER light switch feels like.
You've obviously worked on low force switches before, what kind of things have you done with them?

Just the obvious MX switch modification stuff.
I found out about adjusting actuation force from Tim Tyler.
http://mykeyboard.co.uk/surgery/

[Squeaky Wheel]

You could just use the existing TMK firmware that's out there, and then adapt Plover for Velotype. In principle, that should work.

[trishume]

@Squeaky I can't actually do that because I'm using the Teensy 3.1, which can't run any of the existing keyboard firmwares. Also Plover is designed for dictionary based phonetic layouts whereas Velotype doesn't need a dictionary so it's a totally different architecture. Thanks for the suggestion though.

I'm perfectly happy to do all my own firmware and software development. Programming is actually my main strength, it's hardware design that is the side hobby.

@xman Yah I also got the spring cutting technique from Tim Tyler, but he didn't really give details so I had to do a bunch of experiments to come up with the 1.5->1.0cm cut. Have you just played around with modification or are you actually using a modified keyboard?