midijunki

liner notes: countdown!

note: the following post compiles the accompanying original process write-up accompanying the video into a singular page with additional hyperlinks for ease of access.

my submission for #RSVideoJam3! this is my first video created with the help of reaper. big shout out to ‪@chfr‬ for helping me with the logistics.

usually i don't like to talk about my videos - it's my belief that anything that i would want to say about my work is already embedded into itself - but i did want to try and document this as i went because i think understanding the process of creating this is actually part of the big picture.

to start, i tried to create a problem for myself that was so troublesome that using reaper would be the only practical solution. enter the "spelunker method" (シュペランカー戦法), a specific technique in 音MAD where each note is its own unedited sample. in my regular workflow, this concept would've required me to arrange all the individual notes in ableton, port it to resolve, then rearrange the corresponding videos to their respective notes; with reaper, however, i'm able to arrange both audio and visual at the same time with and export it directly as video output (and it's here where i realize why ytpmvers swear by it so much).

to further raise the ante, i chose to cover the song "countdown" by john coltrane - a legendary jazz piece mostly dominated by a furious saxophone solo over very technical chord progressions. i had already transcribed this solo a few years back and luckily still had a midi file of it available (it's also uploaded to my shared drive, though i don't really expect anyone else to have the need for it.)

i'm not exactly sure why i chose nanahira as my source audio, but i ran into my first major roadblock when her vocal range didn't match that of the tenor saxophone. to fix this, i adjusted the midi file to fit a smaller range while still retaining the intended phrasing - ultimately the compromise seemed to work best between G3-G5. from here i went through her discography to find the notes i needed, which results in the above video. i tried sampling from a variety of eras and focusing on vocal parts that would need minimal processing from uvr. additionally, some samples like C#4 and C5 were stretched out to emphasize the tone of a sample that was not originally meant to be melodic.

i took the previous video and threw it into uvr, using the bs_roformer model. there's lot of documentation out there on installation and use of these, but i followed a recent tutorial from fellow ytpmver なまず.

while this model is generally fantastic at producing clean results, i don't exactly have great hardware...so this model will often take me over an hour to render a 3 minute song to an acapella. this simply isn't practical with ~24 songs to render just for a single note of each, so the secret second reason for the previous video was to create a single object that would only have to be passed through uvr once. (1/27 EDIT: i've discovered that bs_roformer actually overwrites whatever segment size you input in the uvr ui with what's provided in its yaml file; changing it to a value more appropriate to my system fixes the runtime problem without any significant detriment to the output!)

i was worried the collaged audio would create ambiguities about what uvr would determine as an instrument vs a vocal, but the results came out pretty well! i trimmed each clip down to the note i would need to create my "keyboard" for the full song. listening to them in isolation, i had a slight worry about the highest 2 notes as they retained the backing vocals of their original songs, but i also thought it might be interesting to keep them as odd little "harmonics".

to assemble this in reaper i used ePi's midi2item script, which takes a midi track allows you to simply drop a clip in to synchronize to it. the drum variation of this script goes further and creates a track for each midi note, which is absolutely perfect for my use case.

it should be noted that during this entire planning process that this video was still strictly conceptual - i had all the individual notes, but i still had no idea how this would sound as a completed video. when i finished loading all of my clips into the melody and played it back, i was absolutely ecstatic that it really worked! there are some quirks to the clips i chose - C#4, A4 are a bit out of key; C5, D5 start too early resulting in flatness; and the aforementioned noisiness of the F5, G5 - but i actually think these accidents add a human feel to this mechanical approach. other than stretching out some samples for the ending, i tried to keep the clips as unchanged as possible.

the last step was to simply put this to an arrangement and (with whatever time i had remaining) try to spice up this raw output with some editing.

i forgot to mention somewhere at the beginning that part of the additional difficulty of this song choice was that there was absolutely no way that the original recording from 1959 was going to sync to a mechanical clock of a daw, so the next step was to create a musical arrangement around this reaper output. this video presents the entire instrumental without said output, and i've also uploaded another version without the backing vocals to my shared drive.

i kept the editing on both fronts fairly light as i wanted the focus to remain on the spelunker'd output for the majority of the video. the effort begins in earnest at around 1:26 when the actual countdown motif starts (this in itself is very strange for a jazz song, and ends up working like a sort of "drop" if that was even feasible in the late 50s). much like the original song i don't really think this has any place in live settings, but i did want to keep it in tune with the recent trend - and my recent obsession - of making 音MADs that are very club-cognizant.

i want to give to give a shout out to the following songs as reference points / samples for this arrangement (were it not for character limits i could probably go on about the music for days):

top

#liner_notes #ytpmv