Audiosurf Mechanics: Insights, Speculations, and Tests.

[DigitalAssassin]

This was originally intended for the Steam Audiosurf forum, and will probably be posted there some time in the future, but they're not getting back to me about activating my account, so here goes:

What determines the traffic count in an Audiosurf song?  How is the slope generated?  What makes the 3-minute-43-second version of “The Prodigy - Voodoo People (Pendulum Remix)” an absolute nightmare, but when I play the “real” full-length version I purchased from eMusic (5:04 or 5:10 long) I get a much slower track, even though the song sounds about the same?
First, that track:
http://www.youtube.com/watch?v=oPXWUbNYZm0

It’s one of the hardest songs in the game for sure, so I decided to throw it into Audacity and analyze it a little.  It has as very high noise floor, which I determined by splitting the left & right channels, flipping the right one upside-down, then recompiling into a mono track.  This eliminates everything that’s identical in the two channels (and sometimes can extract or eliminate vocals, depending on how well the song was mastered).  I came away with a noisy, hollow-sounding mess, which I proceeded to load into Audiosurf and play.  The track generated was almost identical to the original, in terms of traffic, slope, and 'feel'.  I have no explanation for this.  The peaks were almost eliminated, though the curve stayed fairly similar (remember it’s just one big slope downwards, so this doesn’t mean much).  I guess the average wavelength was about the same, and probably the frequency range.  Is this what is used to calculate the game's track?

Or maybe the traffic and slope is based on the noise floor.  To test this, I generated 25% brown noise in Audacity and mixed it with my version of the Voodoo People track, hoping to come up with the insanity that was the shortened ‘online’ version (which sounded muddier to me).  No such luck:  The track didn’t change at all, except that it sounded like I was playing in a wind tunnel (thanks to the noise I added).


I decided to try generating that 25% brown noise track again and playing it by itself, thinking maybe it got interpreted as nothingness.  I could not have been more wrong.  I can’t possibly describe what I came up with in words, so without further ado, I give you holy grail of impossible tracks:



Obviously it's no fun to listen to, by HOLY CRAP it's absolute insanity.  Some unanswered questions:  Would white/pink noise generate the same thing?  Since white/brown/pink noise is generated randomly, how much does traffic vary from one render to another?  What frequency range generates the most traffic & speed?

The only question all this has answered for me:  Amplitude does not determine intensity.  Dynamic Range might have an effect (big peak minus small peak), but if you amplify a song or negative-amplify a song (making it louder or quieter), you’ll end up with the same (or virtually the same) game-track.

I have no answers to the questions I set out to answer though.  So:  Does anyone have any ideas?  How does the Audiosurf algorithm work?  What’s under the hood?

Thanks! 


PS:  My next test is a pure sine wave which ranges in frequency from 30 to 1000hz or so over a period of 30 seconds or so.  I'll post the results when I'm done.

[Cody900]

Can you upload that Voodoo People song please? Can't find it.

Also, I want that white track. Looks really nice.

EDIT: I think I found it... 5:02 long.... yours is 5:10.

[Pwntastic]

Did you delete the .ash file between trying those various versions of that song? If they have the same id3 tags and the ash file already exists, audiosurf wont regenerate the track and you'll end up with the same thing regardless of how much you changed the actual file

[DigitalAssassin]

Yes, the tracks were recreated each time, and the results consistent (I've repeated them with different file names and id3 tags).

[katonczyk]

Ok, now just generate simple plain tone of e.g. 1khz. Just one tune for like 2 min. You will get the same output as that noise-only file. Track going straight down with tones of blocks.

[Avarice]

Nothing of value to add here, I have no manner of expertise in regard to sound manipulation.

However, I would just like to voice my approval of this thread ^__^  I've wondered what causes some tracks to be so messy when the song itself is not particularly intense.  I correlated it to frequency and pitch, but this thread subject seems a more understanding perspective.

[Lian To]

I wouldn't put so much emphasis on what type of tone or noise causes intensity, when there's the whole beat and intensity variation to consider, too.

I mean if you make a track based on 5% brown noise it might be no different than one with 100% white noise, but if you add a 100% spike to the former, it might be calm for the rest. Sorta thing.

[Melodia]

Did you delete the .ash file between trying those various versions of that song? If they have the same id3 tags and the ash file already exists, audiosurf wont regenerate the track and you'll end up with the same thing regardless of how much you changed the actual file

That's not true at all. Regardless of the way the scores are reported, the ash file IS based on the file, not the same. I've played two seperate tracks with the same Artist - Title (made a thread about it a while back) and it generated a new one each time.

On the flip side, even if you change the ID3 tags, it does NOT generate a new one if you load the same track.


-Lala-

[Rynex]

slowpoke.jpg

Song already developed that is 1013 on Ninja Mono Ironmode.

http://rapidshare.com/files/98475442/sine_of_the_times.wav.html
Plug in some earplugs while you play, or mute the sound.

We still aren't sure why it's so high, but apparently there was people already looking into it.

Just so you know, it has to be a .wav otherwise it loses some of the quality in it. Encoding does have an impact on the song it seems. But yeah, your thread, though informative for some of the newbies here, is late and thus gets slowpoke.jpg.

Also, I tried different sine waves and did slight adjustments and the results were very random. The guy who made the sine wave "flatluigi", seems to of been really lucky. Oh, and the other challenge was to see if the sine wave could be embedded into songs... but I don't think anyone has bothered yet.

[Denmarkian]

I have noticed that when there is a lot of random noise, the "intensity rings" come forth. This is usually when the drumming comprises of a lot of cymbal crashes, as well as when the vocals have pronounced "S", "CH" or "T" sounds. A lot of the j-rock I'm playing have those, especially the j-ska.

What I'm curious about is how much the encoding quality affects the output. I have a pair of songs encoded at 128kbps Mp3 and I have sent them through to be analyzed and I get nothing but a steady purple incline with some crazy music going on in the background. I have also noticed that my FLAC files go through much faster than any of the audio I've imported through iTunes.

From an audio engineer standpoint, I wonder if it's the dynamic range that affects the playability of a song. That would go far to explain how Flac works better than mp3, and why the "burn an audio CD" workaround is a viable solution to when your mp3s won't play in the game.

[BigYeti]

Dynamic Range is my number one suspect...but I have not found any concrete evidence yet.

[cathfaern]

I have many slow song, but audiosurf usually make a very fast track for them... i try to figure out why. That's what i find out:
I think the algorythm of audiosurf try to look for the base of the song. It's usually the drum, and after it find that, it calculate the speed (the slopes) of the song from it. That's why the song goes uphill when the drums fade away.
But how it looks for the base? I think, it tries to find a periodic signal (that's why the sinus wave produce so fast paced track)
Most of the songs have drums, or some kind of base instrument, so in most cases, it works OK.
But in a song, where there's no drum (for example: only singing and piano) it can't find any periodic signals. To be more correct, it have to find something, because it's an algorythm, not a human So it choose the most periodic one, and calculate the speed from that. But in singing, thou you say the syllables relative fast, the song may be slow paced (similar stand for a piano). So in these song, the algorythm have no chance to make the speed of the track right

Other things:
- the hops on the track based on the most significant (the loudest) moments of the base instrument
- the cars based on the loudest moments of the other instruments
(if you play a 2-instrument-song, you can see these clearly)

P.S: sorry for my english

[Trauts]

But in a song, where there's no drum (for example: only singing and piano) it can't find any periodic signals. To be more correct, it have to find something, because it's an algorythm, not a human So it choose the most periodic one, and calculate the speed from that. But in singing, thou you say the syllables relative fast, the song may be slow paced (similar stand for a piano). So in these song, the algorythm have no chance to make the speed of the track right

I'd think this is along the right line of thinking.

If you plug in an mp3 of a comedian, you notice that when he talks, you move, but when he pauses, you pretty much stop moving. Syllables/words causes a lurch forward and a short pause, and longer pauses cause you to pretty much stop moving.

[Bezzend]

Oh, and the other challenge was to see if the sine wave could be embedded into songs... but I don't think anyone has bothered yet.

Has anybody done this yet?
Using Audacity, I've built up a library of 5 minute samples of pure sine waves. 75 of them, covering all frequencies from 1Hz to 75Hz. I ran them through Audiosurf Casual Mono, Ironmode off, and noted the traffic congestion for each. And then plotted a graph of frequency against traffic congestion, which was pretty random and showed no signs [well, not really] of a pattern [although I'd have to do alot more to verify that].

But anyway, that's besides the point, just thought that might interest you.
Okay I was going to write some more stuff here, but to be perfectly honest, I'm clueless. 7Hz is the frequency with the highest traffic congestion out of those below the human hearing register, so does anybody have any idea how to embed that into songs?

[S.]

Lets get back on topic.

I've wondered, where a track as TTFAF creates a fairly fast track, but tracks like Dark Interval (OBOX soundtrack) generate a very steep downhill thing too, even though theres no drum at all.


Some thing to note: If for example you have song that goes downhill, placing it in file with another song, it may go less downhill or even greenish.

Not anything usefull to say, i guess.