It wouldn't be useful for anything. autogen maps would be terrible.
Exactly. Take for example what do you consider rythm? Some want to go more for vocals, others will go for drums.BirdyThePhoenix wrote:
Would be amazing if something like this could be coded well, but the thing is that a machine just can't replace the human element. There are so many different things that make up a beatmap, and a machine wouldn't be able to produce the high quality beatmaps that people can make.
Flanster wrote:
This idea has been floating around for years, its not something new. To be compeltely honest, the amount of work you'd have to do on perfecting such a program would probably be tons more than just mapping something.
Mapping something actually good.Khelly wrote:
Flanster wrote:
This idea has been floating around for years, its not something new. To be compeltely honest, the amount of work you'd have to do on perfecting such a program would probably be tons more than just mapping something.
That's a dumb excuse - you don't make the program to make one map. You use it to make hundreds
Oh well then you should've just said that in the first place.Flanster wrote:
Mapping something actually good.
probably for if you are person with no idea how to map but you want to play the song you like so you can do like this at leastStefan wrote:
The question is: why?
I think manual input of timing would be even slower than actual mappingXiyng wrote:
Of course it's possible. Now getting the program to create maps that are fun to play, that's another thing... It's definitely possible but I don't think it'll be doable pratically ever.
Of course it can be made easier. If you tell the program the timing of each note, you can probably relatively easily make it space them sensibly and maybe even convert some of them into sliders. That's actually an idea... Hmm. If I have the time (which is very doubtful), I might actually want to look into that just a bit. I mean, it doesn't sound too hard.
Maybe Mania too?Rilene wrote:
On Taiko may be possible but the rest, I guess not really a easy thing to program.
Same way computers can beat grandmasters in chess. Is it worth the effort perfecting a machine learning algorithm for this game? Probably not.FlamingPony wrote:
Honestly, hand-made maps are the best and most likely will continue to be the best maps in osu! I just don't see how an "automatic mapping" system could beat something made with human thoughts and actions.
Assuming you have the first beat (aka offset) and know the BPM (both are programmable or you can use user input for better precision), it's not hard to space them correctly. Take any beat you find in your program and move it to the nearest fraction of BPM you have (with 1/16 or 1/8 precision or something). The beats you detect programmatically are already very near from those anyway. I suggest you do a first pass to find sliders/spinners, remove them from your detected beats array then map the remaining beat to hitcircle. That what I do anyway. If you want to get started, you should probably take a look at FFT and onset detection.Xiyng wrote:
Of course it's possible. Now getting the program to create maps that are fun to play, that's another thing... It's definitely possible but I don't think it'll be doable pratically ever.
Of course it can be made easier. If you tell the program the timing of each note, you can probably relatively easily make it space them sensibly and maybe even convert some of them into sliders. That's actually an idea... Hmm. If I have the time (which is very doubtful), I might actually want to look into that just a bit. I mean, it doesn't sound too hard.
b4k4 blue wrote:
I run a Fast Fourier Transform (or FFT for short) on the sound every frame, which separates the sound into an array of frequencies that is analyzed as the song plays. If the user selected frequency is greater than a certain threshold, you have a note.