Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[Armok]

Proposal: train a neural net* to compress and decompress the display until it can get it back identical, feed into it as text, into a short noise-resistant sound with filters to remove anything humans would have a hard time hearing. Then just run the compression part and play the resulting sound as fast as possible. Would take a long time to learn but theoretically should be doable.

* neural nets are often frowned upon, for valid reasons, but I have an intuition saying the result will be more easily interpreted by a human this way even if it solves the training task much less efficiently.

[Rowanas]

I'm not sure how that could be learnt without spending a lot of time setting up single variables and playing the noise, and learning to piece it all together.

White Dwarf: noise
Grey Dwarf: noise
Grass: noise
pickaxe: noise

and so on for every single item, entity and state of being, for a screen full of data, constantly shifting. Can you even get all of the sounds for one timeslice done before the next begins?

[zkline]

I'm going to choose to believe you aren't serious.  It's a fun idea, but really…  My hearing isn't *that* good.  Text to speech is the way to do it. 

[DrKillPatient]

Zack, I'm not sure how well DF would perform this way, but you could use a live CD with Linux on it; something lightweight like Knoppix might do the trick. If the live CD can read and mount your hard drive, you can download the Linux version of DF onto your Mac's internal hard drive, and run Linux from the CD. Then you'd mount your hard drive from the live CD and run DF from there. Linux definitely has text mode, I'm running it right now and I've seen it do ncurses output (although I don't use it regularly).

One problem is that DF won't have any sound when it's run on Linux without a little fiddling around, although I think you might want to use your own sounds rather than listen to DF's music, which isn't particularly necessary, and would likely get in the way.

And by the way, setting up a live CD is fairly easy: you just download the ISO for your language, and either a small CD image or a larger DVD image (depending on what you have at the moment, either will work fine for this purpose), and burn it to a disk. Then you can just restart your computer while the CD is in the drive and it should start up.

[zkline]

That's a great idea.  THe issue I have with it has a lot to do with Linux screen reader limitations, unfortunately.  THe best Linux screen reader for the console isn't Unicode aware.  A lot of DF characters, dwarves for instance, are Unicode symbols which the program uniformly calls "null."  THe OS X screen reader has the advantage of being Unicode-aware.
Thanks for the suggestion,
Zack.

[DrKillPatient]

In that case, perhaps your VMware idea would still work. Can the screen reader look at the text displayed in another operating system running that way?

[zkline]

Nope, unfortunately it can't.  If it were that good, I'd be very surprised indeed.  Too bad really, it would make things simpler.
Best,
Zack.

[Jurph]

So your screen-reader -- which for our purposes is your primary interface -- works best in OS X, but DF won't render as text in OS X.  Do we know yet whether your screen reader can accurately read info from DFTerm?

Also, Zack, have you listened to the sounds from SoundSense?  It plays sounds based on things being reported in the game's announcements log.  Since DF is a very interrupt-driven game, the idea of having events actually flow out to you instead of you having to mine them from the entire state of the game seems useful.  Having them flow out as words instead of sounds would require you or us to record a sound library, but it runs in Java so OS X should be able to make it work.

[zkline]

Yes, you've understood the situation correctly.  DF-Term seems to read correctly so far.  As far as SoundSense goes, that seems an ideal solution, or at least something definitely worth investigating.  Text rendering in OS X is also a great place to start.
Thanks,
Zack.

[Kurskovich]

you could get DFterm and connect to the game hosted on the same computer, I don't see the problem with this if DFterm can be read by your reader

[Jurph]

you could get DFterm and connect to the game hosted on the same computer, I don't see the problem with this if DFterm can be read by your reader

Kursk, DFTerm requires you to be running the game in text mode in order to transmit the terminal.  It's like saying if he had four good tires he could drive to the tire place and get his flat fixed.

[zkline]

GOod news, everybody.  I was able to connect to the public DF server and generate a world, embark, and read the symbols just fine.  The biggest problem, and unfortunately this is a biggy, is the inability to track my position horizontally.  I can move up and down line by line, but unlike with my Linux screen reader if I do that my horizontal position isn't preserved.  This makes reviewing things in rogue like's difficult at best.  I've gone ahead and put in a request at Apple for this feature.  In the mean time, I'm sure there are things we can do.
Best,
Zack.

[Armok]

I'm not sure how that could be learnt without spending a lot of time setting up single variables and playing the noise, and learning to piece it all together.

White Dwarf: noise
Grey Dwarf: noise
Grass: noise
pickaxe: noise

and so on for every single item, entity and state of being, for a screen full of data, constantly shifting. Can you even get all of the sounds for one timeslice done before the next begins?

Which is why you need to use some more advanced forms of compression, that only describe larger-scale structures.

I'm going to choose to believe you aren't serious.  It's a fun idea, but really…  My hearing isn't *that* good.  Text to speech is the way to do it.

You don't need any especially good hearing, just lots and lots of practice. And having a lot of unused neurons in your visual cortex should help much more with this than other kinds of hearing.

[zkline]

Let's pick one approach or a combination of parallel approaches, as Jurph has outlined.  RIght now i think the most serious problem is the difficulty of tracking vertical moment, as I indicated.  To illustrate what I mean, say you have a 25x80 grid, like a typical text mode screen.  If I were at (5,5) on that grid and moved my review cursor downwards, the Linux screen reader would place me at (5,6).  All other programs I've tried would put me at (1,6).  This is hardly ideal for roguelikes.
Perhaps a separate program could be written to allow this sort of tracking, I don't know, but at this point the barrier to entry is quite high.
Best,
Zack.

[cameron]

one thing that might make things a bit easier is if you keep your embark size small enough that you can see everything without moving the screen around, that way static features will always be in the same location on your screen.