#audio spectrogram

[graphicnoises]

Wheezy/whistly laughter is so cool looking on spectrogram.

[graphicnoises]

If you've ever wondered how those automatic audio-to-(bad)-sheet-music programs work this is a good chunk of the kind of stuff that goes on in the back end.

Each of these flat lines is long enough to be easily and accurately identified as a specific frequency and duration. Looking at the axes and cross-referencing with Wikipedia, we can quickly identify that faint line that's constant up across the rest as C6 (1046Hz), the lowest line as C2 (65Hz), and strong lines at 98 and 198Hz, which are G2 and G3 respectively. That's a solid indication, absent any other information on the song itself, that it's set in C.

However, we have additional information in this graph. Note the time axis in seconds across the bottom. This could enable a particularly diligent investigator (or, more likely, a program) to track the points in time when notes begin and end, and attempt to calculate common denominators that predict those notes. In this case, a researcher with particular time to kill could mark the individual duration of every single note, and use the running GCD as a point estimate of base note length from which to calculate an estimate of the BPM.

If you want to get an idea on how Fourier Transforms themselves (and by extension, the FFT algorithm) work, I'd recommend looking up a 3Blue1Brown video on them, their work is very good.

Song: Beekeper - Keaton Henson (Spotify)

[graphicnoises]

Choppy distortion effects look so pretty. It's like someone went over the audio track itself with a rake. In the best possible way.

Song: Persona Non Grata - Whale Bones (Spotify)

[graphicnoises]

1: I forget the music-y term for it but that fade is so pretty both audibly and graphically (it's accurately reproduced, I'll say that)

2: This graph legitimately makes it so much easier to hear the lower frequency tones such as the bassier parts of the drumbeats. I'll make a point of including something with an actual low bassline later.

Song: The Wolves - JJ and the Pillars (Spotify)

[graphicnoises]

Strong vibratos are so pretty. I guess it makes sense they'd be pretty on spectrogram as well.

You can also see how it comes out so much more strongly near the ends of held notes, when the singer has stopped holding the note as tightly and loosens up.

Song: I Scare You - Roaring Girl Cabaret (Spotify)

[happynuke]

Visit Original Link

This is the result of my meddling with an audio spectrograph and Pokemon Red. I chose Lavender town for obvious reasons. I was very surprised by the end result. It sounds so familiar to the original tune, yet it's changed enough to be completely different. It makes no sense. It only seemed to want to produce 30 seconds of this before spewing out random sounds.

This is the spectrograph I used to 'redraw' the music. It's a direct reading from the Lavender town track. Audio Paint recreated what it thought it read from the image. It's messed up, but still retains its 'creepiness'.

[happynuke]

I'm not really sure what the result will be...

My next project has had a drastic change of theme. It's almost pointless but it sounds like a bit of fun.

I'm going to use an audio spectrogram to analyze tracks from games notorious for scary things/creepypastas. This includes Pokemon Red, Shining Force II (that freaky boss battle music!!) and any other soundtracks from actual frightening games that I can think of.

Suggestions are welcome! I will post the results in image form tomorrow.