Japanese input’s core part is less the direct input and more the conversion to the target character (which relies on the IME, so we’re not far from the “chorded” mechanism of this keyboard)
For instance there is a “kana” mode that let’s you direct input each kana with one keystroke only, which is way faster than romaji input, but is ignored by 99% of the population because the ROI on learning it is extremely low as you’re still converting to kanji at the end. To your point, a ten-key style input on desktop would probably be slower than kana input.
Is Kana input really that much faster than Romaji? I've almost never seen it used (presumably because you have to learn romaji input to type in English anyway, so why learn a second input method), so can't speak to its speed.
However, is it possible that ten-key style input could actually be faster than Kana input? After all, direct Kana input in Japanese is kind of equivalent to Roman input in English in the sense of one key = one character, but this keyboard claims to be faster than direct input due to not needing to move your fingers around. Couldn't the same thing be true for Japanese? In fact, for Kana input the problem of finger travel is even worse due to more keys = longer travel.
> this keyboard claims to be faster than direct input due to not needing to move your fingers around
That would perhaps work if you remapped the movement to not be linked to romaji, but direct kana mappings. Then it's not just 26 keys combination for the alphabet alone, but 46+ more arrangement, and more for ° and ". With 10 fingers it feels like hell just for standard text input (to note, you can only use 9 for combination as you need a transformation key, and I guess you also need backspace and arrow keys ?), and you add to that punctuation etc.
I'm not sure there can be an actual human being effectively typing faster with this...
> That would perhaps work if you remapped the movement to not be linked to romaji, but direct kana mappings.
Yes, I picture the same 12 key basic input mapping as on smart phones. The CharaCorder has enough joysticks, but looking again at their website it doesn't look like it would work—seems like each joystick only supports 4 directions?... too bad.
For instance there is a “kana” mode that let’s you direct input each kana with one keystroke only, which is way faster than romaji input, but is ignored by 99% of the population because the ROI on learning it is extremely low as you’re still converting to kanji at the end. To your point, a ten-key style input on desktop would probably be slower than kana input.