The Recluse Report - September 2024 Part 2

30 September, 2024. 1,689 words.

I’ve spent a lot of time this cycle sitting around in a stuporous malaise, not writing, not programming, not gaming, not making videos or music.

Gaming

I’ve only played Dragon’s Dogma 2 a couple of times this cycle, otherwise nothing.

Media Consumption

Regularly consuming: Dimension 20’s Time Quangle (on Dropout.tv), Taskmaster Series 18 (on YouTube), Guy Montgomery’s Guy Mont Spelling Bee NZ Season 2 (on VPN), Glass Cannon’s Blood of the Wild Podcast (on Google), The Glass Cannon podcast (on YouTube), Daily Show clips (on YouTube), The Weekly Show podcast (on YouTube), The Rest Is Entertainment podcast (on YouTube), Bill Maher’s Real Time (on Max, but usually just the Monologue and New Rules).

Background consumption*: Taskmaster U.K. rewatches on YouTube because I spent a lot of time lying in bed resting my stupid back.

* Except in situations where I actually need to concentrate or when I’m recording something myself, I almost always have something playing in the background 24/7.

Wings of Pegasus Videos

I don’t know how one of these videos got onto my recommendations, but I fell down a rabbit hole of watching YouTube videos of this guy Fil of Wings of Pegasus, a channel I’d never heard of before, analyzing music videos and performances to see if they’re using auto-tune or pitch correction or lip sync.

This is apparently a controversial subject and he attracts a lot of criticism (the recipe for all successful content creators). As a person who has a certain amount of audio engineering knowledge, I can report that he sounds completely authentic and knowledgeable of the craft to me, though he does tend to descend into random or repetitive word salad a lot. It feels like he spends 10 minutes explaining things that only need 1 minute to explain. And there is a strong undercurrent of “this is how good I am at music so make sure to check out the stuff I sell” vibe to it, along with the usual big content creator shouty headlines and whatnot.

That aside, the point is, there is apparently a whole world of people out there who are into exposing musicians for using pitch correction (and/or debunking that musicians have used pitch correction). There is plenty to criticize about the modern music industry, but analyzing pitch correction seems like a weirdly specific hobby to get into.

My 2 cents (har-har! get it? pitch is measured in cents!) is: Yeah, I mean obviously everybody on the Spotify Top 100 and in those Huge Arena Shows are using auto-tune and pitch correction and lip syncing. Music is way too big of a business not to. You really think The Eagles are going to cancel a billion-dollar show because Don Henley gets a cold? I think not.

The other side-effect of watching these videos is that I found a bunch of tools to isolate vocals and other instrument tracks. Which leads me to…

Media Production

When my NAS crashed last year, I had backups of most everything important, except for audio projects created between around 2020 to 2023 (and there weren’t very many). However I lost immediate access to a ton of the source files for my archived music projects.

So I embarked on another journey of copying backups of old audio files and projects from CDROMs and hard drives onto my NAS. I bought a gizmo that lets me plug in old IDE hard drives into a USB port so I can quickly copy the files from the drive, without having to setup any old computers. It’s a pretty cool gadget.

Between back-resting sessions, I’ve spent some amount of time with REAPER and FL Studio this cycle, trying to setup (or actually re-setup) working environments in which to load all my old music. I started a new project in FL Studio using some audio loops I got in a Humble bundle to try to remember how it works.

The Lost ADAT Tapes

Long story short: I wrote and recorded a lot of music during the 1990s, because reasons, mainly revolving around the youthful misconception that art would continue to be valued in the age of the Internet.

I have digital copies of most of the source material for that music I recorded, but not everything.

After moving many times, I cannot locate any of the original ADAT tapes of my earliest recordings anymore. (It’s impossible for me to imagine that I threw away ADAT master tapes, even if I didn’t have any way to play them, and I certainly don’t remember doing it, but I’ve turned my house upside down many times in the last few years and I just can’t find them.)

So there is a subset of my music from about 1993-1995 where I only have the final mixdowns on CDs and DAT tapes. This is infuriating, because I’d love to remix some of those songs using modern technology, but it’s basically impossible without re-recording them from scratch. Which is quite a chore because I can’t play or sing even half as well as I could back in 1993-1995. (Luckily the tools for fixing bad playing are practically limitless now.)

That’s all background for the real topic of this section: I found this YouTube channel that isolates vocal tracks and analyzes them (see above), and I wondered if I could make use of that myself to recover some of my lost tracks. Apparently this is yet another new thing AI (or “AI”) is really useful for. (See also: The last Beatles song.)

So I looked for some AI tools for isolating vocals. They’re plentiful online. I picked this site called Lalal.ai, where you can upload a song, and it will attempt to separate out not only the vocals, but also the drums, guitars, or bass.

I gave it one of the songs I don’t have individual tracks for, and it worked. In an imperfect way, at least. It’s nowhere near as good as finding the original ADAT tape and transferring the tracks directly. (That particular tape, however, was mangled by my ancient ADAT deck, so it might be completely impossible to recover even if I could find the tape again.) But it’s better than nothing.

Here’s an example of it in action:

The original mix, which I remastered in 2010:

AI-isolated drums (from a Boss DR-660 drum machine sequenced on an Amiga):

AI-isolated bass guitar (a real bass guitar played by me):

AI-isolated guitars (a real strat played by me):

AI-isolated vocal (a much younger me):

By extraordinary luck, I actually did find two of the original seven tracks from that song in WAV files (I have no idea why). One of them is the vocal (the other is the lead guitar). So you can compare how AI did against the original:

From my own casual comparison I can report there are numerous artifacts in the AI-isolated vocal that make it pretty much useless garbage from an audio engineering perspective, but it’s fairly miraculous that it was able to get anything at all from just a handful of mouse clicks.

Lalal.ai is pretty good, but I’m also interested in trying out Spleeter, which is a less user-friendly implementation of the same technology, this time run from a command line. It has the distinct advantage (allegedly) that you can feed it your own training data. I have tons of my own isolated vocal tracks I can feed it, and isolated guitar tracks, etc., which should theoretically improve its ability to isolate my own voice from a mix.

The moral of this story is that there might be some hope for remixing those songs after all, thanks to that evil AI thing that we all hate, but still find uses for every day. If only I had the time and energy to spend on remixing old songs. I should probably just give up and write new songs.

Home Development

I’ve been taking a break from blog comments to rest my back. Luckily I was able to get them into a semi-working state before stopping. Surprisingly, the current implementation is such that I’ve received exactly zero attempted spam comments since deploying it.

Day Job

This sprint was a bit slow because there was a moratorium on internal deployments that kept me from making much progress.

Health and Wellness

If I believed in biorhythms this would have been a time period where my physical biorhythm graphs were way down in the valley part. My back and neck and shoulder blade area have been bothering me greatly, which cascades into a lot of time spent doing almost nothing productive. Every cough or sneeze feels like a re-injury. It was almost impossible to spend time sitting at a computer using a keyboard or mouse without feeling pain.

It’s been doubly frustrating because I’m a person who has a constant stream of ideas for projects that I want to work on or things I want to do, pretty much all day every day from the moment I wake up until the moment I fall asleep, and yet, I have to table them and hope I can work on them later. I had to spend almost 100% of my weekends resting up from the exertion of trying to sit and do programming work during the week. Some people out there probably understand what that’s like. If not, then I can report that it sucks.

World Context

As yet I don’t know what to replace TweetDeck with for news.

There was another honest-to-gosh assassination attempt on Donald Trump in Florida, though this one was foiled before he was in any danger.
Israel has been escalating attacks on (Iran-backed) Hezbollah in Lebanon, using air strikes and–I promise I’m not making this up–exploding pagers.
Ongoing Trainwrecks of the Year: 2024 Presidential Election, War in Israel (since 10/2023), Sudanese Civil War (since 4/2023), War in Ukraine (since 2/2022).
Periodically I check the status of the ongoing trainwrecks and it turns out the Nigerian Crisis (since 7/2023) as a result of the 2023 coup “officially” ended in February, though U.S. troops only just completed their withdrawal from Niger.
Celebrity Deaths: Dame Maggie Smith (actor), Kris Kristofferson (musician).

Bye!