I’ve been thinking about how to type faster than 110wpm, and digging into the specific factors that I could improve. In particular, I wanted to get a sense of:

• my theoretical top speed
• whether alternates or rolls are better for me
• how quickly I can twitch, measured by single-key repeats or two-key alternations

By using totally artificial typing tests (ex: type “thththth…”) instead of word-based ones, I can explore the relationships between character combinations and speed without worrying about hitting SPC, sounding out words, correcting errors, and so on. Since I can do the tests in short sprints, I can rest enough in between to minimize my risk of RSI.

Using Emacs to test my raw typing speed

I haven’t come across an online typing test that gives the kind of stats I want, or even a per-character or digram breakdown. I thought about writing a Javascript-based typing timer, but I figured it would be less work to cajole Emacs into measuring what I wanted. Here’s the code:

(defun sacha/timer-go ()
  "Quick keyboard timer."
  (interactive)
  (insert "GO\n")
  (run-with-timer 3 nil (lambda () (insert "\n")))  ; for warmup
  (run-with-timer 15 nil (lambda () ; 12 seconds + the 3-second warmup
                           (let ((col (- (point) (line-beginning-position))))
                             (insert (format " | %d | \n" col)))
                           )))
(local-set-key (kbd "<f7>") 'sacha/timer-go)

This prints “GO” to show you that it’s running. You have three seconds to warm up, so you don’t have to worry about wasting any milliseconds after M-x sacha/timer-go (or F7, the keyboard shortcut I bound mine to). After the warmup, Emacs adds a newline and the “race” is on. There’s a 12 second period of actual typing, and then Emacs adds the number of characters you typed. When you see that, you can stop.

Twelve seconds is a useful number for estimating typing speed because the conversion from characters per minute (CPM) to words per minute (WPM) usually uses a factor of 5: CPM / 5 = WPM. So the number of characters you can type in 60 seconds / 5 is probably the number of “words” you could type in a minute.

Note: L and R refer to left and right hand. I’ve also numbered the fingers with 1 being the thumb and 5 being the pinky. The patterns I used are based on a Dvorak keyboard, but that doesn’t matter as much. You can probably figure out what the equivalent patterns are on your preferred keyboard layout.

Limitations: I didn’t do any special calculations to deal with errors (there were many doubling or transposition errors multi-character sequences), so the actual CPM will be lower. Also, repeated character sequences are definitely not normal and have quirks of their own. It’s interesting to establish the range and see the kinds of errors that show up when I go faster than I’m comfortable with, though.

Pure speed

Key description	Pattern	Estimated WPM based on CPM/5
keyboard mashing	-none- (mashing)	379
keyboard mashing	-none- (mashing)	379
keyboard mashing	-none- (mashing)	354
R side mashing	-none- (mashing)	245
L side mashing	-none- (mashing)	217

If you don’t care what you’re typing, it’s easy to type quickly. This is just about how fast my hands go if I don’t have to think about which finger to activate. This mostly ended up as alternating left- and right-hand rolls (ex: aoeusntoahuesnto). Because I didn’t have to precisely alternate, two-handed mashing resulted in more characters than one-handed mashing. Interestingly, my right hand is slightly faster than my left.

Alternates versus rolls

4-key combinations

Key description	Pattern	Estimated WPM based on CPM/5
R-side 4-key roll	snthsnth	232
L-side 4-key roll	aoeuaoue	201
L 3 & 2, R 3 & 2	eutheuth	164

3-key combinations

Key description	Pattern	Estimated WPM based on CPM/5
R 3 & 2, L 3	thethe	187
L 5, R 4 & 2	andand	184
R 3 & 2, L 3	thethe	182
roll R 3	nthnth	176
R 3 & 2, L 3	thethe	170
roll L 3	oeuoue	166
roll L 3	oeuoeu	164
R 3 & 2, L 3	thethe	159
roll R 3	nthnth	152
R 3, L 4 & 3	toetoe	140

I expected rolls to be faster than alternates, but it turns out that alternating works out fine too (“the” and “and” on a Dvorak keyboard). Same-hand rolls had fewer errors than alternates, though – timing can be tricky when doing high-speed repeats. That can be partially handled by autocorrecting “teh” to “the” and similar transpositions. I use an AutoHotkey-based autocorrect script, but it screws up the typing tests I like, so I can’t take advantage of it then.

A roll-optimized keyboard layout might be more effective. 3- and 4-character rolls like the ones I tested aren’t that common in actual typing, but it might be possible to find keyboard layouts that are better-optimized for the languages I use. I’ve read that Arensito, Capewell, and Colemak focus more on rolls and alternating rolls, so they might be worth a look.

Two-character pairs

Key description	Pattern	Estimated WPM based on CPM/5
alt L and R 1	uhuh	139
L 5, R 5	asas	137
R 2 & 3	chch	135
R 2 & 3	thth	134
L 2, R 3	tutu	130
R 3, L 4	toto	129
L 2, R 2	uhuh	128
R 1 & 5	xsxs	126
L 2 & 3	eueu	124
R 2 and 5	shsh	115

Two-character patterns are slower than three-character patterns, probably indicating that there’s a small delay as I think about repeating things. Alternates and same-hand two-character pairs seem to work okay. Even for same-hand two-character pairs, I get the occasional doubling or transposition error.

Single-finger twitching

Key description	Pattern	Estimated WPM based on CPM/5
R 2	hhhh	79
R 3	tttt	76
R 1	mmmm	75
R 4	nnnn	74
L 2	uuuu	73
R 5	ssss	71
L 3	eeee	71
L 4	oooo	65
L 1	kkkk	64
L 5	aaaa	61

Single-finger keypresses (no automatic repeats) are slow. Good thing I don’t have to do them that often. If this represents the speed at which I can send an impulse to my finger and have it do something, this might be a limiting factor for my typing speed, which is compensated for by alternates and rolls.

Three characters with repositioning

Key description	Pattern	Estimated WPM based on CPM/5
R 3, L 2, L 2	cupcup	67
R 3, L 5, R 3	catcat	66
R 2, L 4, R 2	dogdog	64

Moving my fingers takes time too. Also, did you know that there are typing equivalents of tongue-twisters? I can’t type “ranranranran…” a long time without it turning into rna and other permutations. Maybe my brain gets hiccups.

Interrupted combinations

Key description	Pattern	Estimated WPM based on CPM/5
R 4, L 4, R 3	notnot	63
L 4, R 4, L 3	oneone	57
L 5, R 4, L 3	areare	55

Alternating hands is actually pretty tough if you have to care about timing. Oddly, this is slower than repositioning. Maybe it’s because the repositioning helps me remember where I am in the word when I’m repeating it, so natural typing will be a different case.

Wrap-up

Chunking seems to make a big difference for me. 4-character combinations tend to beat 3-character combinations and those tend to beat 2-character combinations, unless there’s some timing involved. Common combinations (the, and) are easier to type. If I can get better at chunking words into syllables, that might help. The most common digraphs are TH, HE, AN, IN, ER, ON, RE, ED, ND, HA, AT, EN, ES, OF, NT, EA, TI, TO, IO, LE, IS, OU, AR, AS, DE, RT, and VE (source), so that might be good to look at next.

Twitching or moving individual fingers are slow operations, so being able to “look ahead” and move my fingers to the right spots while I’m typing the first few characters helps. Muscle memory also helps minimize errors. Also, maybe finger dexterity and agility exercises?

I’m probably in the region of Diminishing Returns here. I could spend hours inching up my typing speed… or I could spend that time doing other things. Now that I’ve identified specific areas to look into, though, I might be able to set up exercises to take advantage of interstitial time. For example, while I’m reading a book, I could do finger dexterity exercises (pausing, of course, if I feel any hint of strain – I’d like to avoid RSI if I can).

On another note, testing my theoretical speed in this way reminded me a little of how we used to play Decathlon on the computer as kids. (Was it Microsoft Decathlon? The screenshots look familiar…) Somehow our keyboard survived the rampage back then. =)

Next steps

Because alternation can lead to typing errors or slowness for me, I might look into Colemak, which optimizes for single-hand rolls. Still, I’m pretty happy with Dvorak, and the Colemak FAQ warns that the switch might not be worth it. Another thing I’m looking into is Plover, which lets you do stenography using a regular keyboard. My laptop keyboard can’t easily do some of the combinations and I’m more visual than phonetic when it comes to words, so it might be a challenge to learn.

The easiest win will probably come from training my speech recognition software to recognize my words more accurately. I’ve been dictating book notes to my computer. This is great because it reinforces the key points of the book in my memory, trains the computer, and helps me practice clear diction. I’ve gotten to the point of using speech recognition to take notes during my first pass through a book, editing after each paragraph. I feel that the accuracy is gradually improving. I make fewer edits as I learn how to speak the way the computer wants me to and I teach the computer to understand the way I speak.

Besides, an average of 107 wpm on Dvorak is fast enough to let me get words out of my head and onto my computer, and I can focus on what I want to say instead of how to type.  There’s plenty more to learn about how to write efficiently. Time to go back to David Fryxell’s How to Write Fast (While Writing Well)! So it’s interesting to dig into what my rate-limiting factors are when it comes to typing faster, but it’s even better to focus on how I can think faster (although speech recognition will still be useful for the benefits mentioned above).

Image credits: Keyboard with time (Cienpies Design, Shutterstock)