I'll admit that I went to the page skeptical, expecting another "let's see how some arbitrary mathematical function turned into a song sounds," but I was pleasantly surprised that some of these sound decent.
It makes me wonder: what if we take existing songs, try to find space-filling curves that explain them, and then look for different paths through those curves? In other words: would it be easy / possible to parameterize existing songs as curves, so that we can find which subset of the space of space-filling curves is actually interesting to human ears?
Well that's interesting. The fast-playing 2d curve (the fourth audio file, under "Meander fast track") sounds a lot like a solo, a taksim, in an oriental scale - Ussak, perhaps, or Segah, although the tuning is western. It's like music made out of a heavily stylised and simplified arabesque. So cool.
The synth cutting off and glitching are so irritating! Seems like a lot of work to put into something to leave really jarring artifacts all over. If you ever create or edit audio, all tracks and samples should start and end on the zero crossing. Otherwise the jump "back" to zero introduces lots of noise. Sticks out like a sore thumb on a spectograph
Reminds me of the postscript to this Ben Krasnow video (from about 13:45) which discusses an analogous effect in a circuit using a capacitor. https://youtu.be/xyMH8wKK-Ag
on the same topic, i been also curious of something and wonder if anyone have the answer to this.
lets say i have a track playing just one note (to be specific, a frequency measured in hertz, lets say 528) for one minute and we want this to be as pure sounding as possible.
what is the best approach to do this?
in addition, if the audio file is in WAV and we convert them to mp3, do we still lose a lot of quality even though we are just playing same note for 1 minute?
how cam we achieve playing the purest sound without music file taking up too much space?
In theory, an 1056 Hz 8-bit WAV file filled with alternating +127 and -127 samples will perfectly reproduce a 528 Hz tone. That's about 60kB in size, and a traditional LZW compression (I.e. Zip) will work quite well due to the repetition.
Not if the sound driver is doing its job properly -- it's generally bad form to add harmonics above the Nyquist frequency[1] of the input file. If you do have to deal with a buggy sound driver, you can use any small multiple of the tone frequency as your sampling rate and still have efficient LZW compression.
Actual sawtooth on the left (done at higher bit rate), tone on right is 2khz tone at 8khz sample rate (audacity won't generate tone right at nyquist). Looks like a triangle wave in audacity, but the tone is fairly pure with just a whisper of harmonics, and those are probably from the speaker.
That won't be possible given the current mapping procedure. It imposes a certain set of constraints on the space of possible tunes that are not obeyed by any of Bach's works: A fixed number of voices (number of notes that can be played simultaneously), greater than 1, with all voices always playing some note (i.e., there are no rests).
If a piece of music does not obey those constraints, it cannot be back-transformed into a continuous curve in a fixed number of dimensions. You would need to come up with some other, more complex, method of mapping curves to music in order to handle arbitrary musical compositions.
It makes me wonder: what if we take existing songs, try to find space-filling curves that explain them, and then look for different paths through those curves? In other words: would it be easy / possible to parameterize existing songs as curves, so that we can find which subset of the space of space-filling curves is actually interesting to human ears?