Help with organizing soundtracks

[Watterson]

Watterson: what I would do, is probably have tempo separate from moods/genres. Also for tempo I would prefer using BPM (at least under the hood).

 

I kind of cut down what I was saying, since as I was re-reading my post I noticed I had gone a little too mad sciencey with things, but I guess in doing that I cut out some much needed context. I also appear to have made mistakes when bitesizing myself, as "fast-tempo" and "slow-tempo" were meant to say 'upbeat' and 'gentle' respectively. I actually only just now caught that when re-reading my previous post to make sure I wasn't about to correct you with incorrect information. This likely happened because of me chopping bits and moving them around haphazardly as I'm want to do sometimes. I have a serious issue with digressions, in case it wasn't already apparent. Believe it or not, you're currently reading something that actually has a lot of other digressions cut out. The fun part about this is any sentence that isn't a digression that doesn't have a digression immediately leading it or immediately following it is an easy-to-spot signifier of where these incisions occurred.

 

Anyway for 'upbeat but happy' an example would be https://www.youtube.com/watch?v=rO_R5puXs6E. An example for 'gentle but happy' would be https://www.youtube.com/watch?v=zSif77IVQdY. These are both songs that make me feel happiness but anybody can see how they're also entirely different. This is where my perplexity came, as I never found a great way to weigh the 'nature' vs the 'general mood' of a song. Then I decided to make it easier on myself by removing the arbitrary 2-tone mood limit and have an infinite amount of tags that are weighed in their accuracy by myself through the application. But that raised the fidelity so high that it became a worse option than manually hand-picking playlists and evolving them over a very long period of time naturally, since it wasn't much faster and wasn't much simpler. In fact the only thing it did better was incorporate new songs into new playlists faster, but with much less satisfying results. See I had this exclusion system. The idea at first was to ensure that certain types of songs (Like incredibly melancholic songs) would be excluded from a playlist labeled as "incredibly happy." But that just added so much more time required to tag a new song and each generated playlist then required even more maintenance. And even then that ended up more or less just turning the application into a high school with heavy stereotypical cliques. Pro Era would never be caught dead sitting at Ana Gasteyer's lunch table. Not unless you gagged them with a spoon first. So that meant whenever a particularly somber STEEZ track like

https://www.youtube.com/watch?v=B7CN6gsHIz4 was put in the same playlist as https://www.youtube.com/watch?v=pl15PlIXHIk because the principal wanted to promote camaraderie, it left this real uncomfortable tension in the air. Like a fight was about to happen, but since it's a highly organized high school it means a fight wasn't going to happen, but it would certainly make the two groups feel much more awkwardly hostile in future interactions with eachother.

 

So this is why I just stick to organizing playlists in folders, copy/pasting to my heart's content. It's also why I'd like to know exactly what Ross does, and exactly what he's looking for. Not because I'm some super niche creep, though there's no guarantee I'm not. But if the problem he's having is simply a convenient place to store his songs and randomize a playlist without having to drop duplicates into random subfolders of random subfolders of random subfolders then that's a pretty simple fix. If the issue is more esoteric, then it's much more complex, but it could be very fun to discuss, though I don't see a solution springing from the discussion. From what I gather he already has something that's suiting him well enough not to swear off music altogether, but is just looking for more practicality. It could prove a neat exercise to pool the community's brainpower to wax philosophic about this issue, to see if we can come up with a better philosophy behind defining and generating playlists, but I'm sure there are people at Apple that have much more resources and that know much more about the subject yet, to my knowledge, nothing earth-shattering has come out of them in this department. That's why I don't see discussion doing much.

 

But hey, hi by the way, don't know if I've introduced myself, call me Watterson, if Ross already has a clearly reasoned out end-goal that he'll be able to easily tell when he's reached, then I'm willing to enlist myself to help. At least with the logistics end of it.

[Qwx]

Hey there Watterson . Sorry for real slow response time.

 

As I mentioned earlier, I don't like files in folders idea just because it's kind of messy, and I'm a bit of a pedant . In a sense what you're doing with folders is an algorithm and it could be implemented in some software. Have a software that has virtual folders with subfolders where you drag or copy stuff around and you have the same thing. Could you give some examples of your folder/subfolder structure? What are the criteria for classification music at different folder levels? Couldn't this be implemented by simple playlists?

 

But that raised the fidelity so high that it became a worse option than manually hand-picking playlists and evolving them over a very long period of time naturally, since it wasn't much faster and wasn't much simpler. In fact the only thing it did better was incorporate new songs into new playlists faster, but with much less satisfying results. See I had this exclusion system. The idea at first was to ensure that certain types of songs (Like incredibly melancholic songs) would be excluded from a playlist labeled as "incredibly happy." But that just added so much more time required to tag a new song and each generated playlist then required even more maintenance.

Well, if we would make tag based system where themes were defined by moods, we could easily exclude certain tags (like melancholy) from certain themes. At the same time, I can see it decreases the control over individual songs in every "theme".

 

but I'm sure there are people at Apple that have much more resources and that know much more about the subject yet, to my knowledge, nothing earth-shattering has come out of them in this department.

Markets always have gaps to fill in. Products cater to certain needs. One of the design ideas behind Apple is - less customization for the end user. Customization adds complexity, which actually worsens user experience. I think what you and Ross are trying to do is quite niche and might not be addressed at all. Or addressed inadequately for you (there were several software solutions posted that did mood thing). If we can figure out what would work, we could implement it .

[Watterson]

Well the 'structure' of my folders is very chaotic to an outsider without my brain. For example I have a folder named "GREAT MUSIC" which is just all southern plantation work songs. I know what this is and what it means because I remember the enthusiasm I had when I was creating the playlist. It doesn't exclude other songs from being great, and doesn't encompass the entire spectrum of songs I would consider great.

 

Basically I just intuit all my needs, naming and categorizing things on impulse, because it's the most effective way of remembering the idea/mood the folder was conceptualized with. The process is just me finding I want to listen to something for a specific reason (Like concentrating on some memory game) but I don't have a playlist that tickles the part of my brain that helps me concentrate on a memory game, so I experiment and eventually come up with a folder I end up naming "Thought provoking." There's another folder called "Concentration," but its purpose is specifically for when I find myself a little too scatter-brained and mentally disorganized. These 2 folders share a couple songs and even a similar concept, but one can't quite substitute for the other.

 

Making a program that does this virtually is practically a waste of time for my specific needs, since Windows Explorer launches on startup and Windows Media Player instantly loads the playlist when I hit ctrl-A and Enter. To program something that will skip this and automatically generate playlists to suit my very specific needs I'd need to have a deeper understanding of my own personal thought processes when editing/creating a playlist in order to predict what results I'd want most when I apply 3 abstract concepts to a desired playlist, and then figure out how to define these abstract concepts in 1s and 0s to fit my definition of them and my prediction of how they'd mesh together. So instead I just intuit everything by mashing songs that fit my need together in folders. That's why I hope Ross has a clearer envisioned end-goal in mind than I do for myself, because if it's more abstract or esoteric I'll struggle (As my track record has shown). If it isn't, then I may be of service. For instance, if he knows how to define his musical taste into a mathematical algorithm and just needs an application programmed to fit his specific needs then I'm in.

[Qwx]

I can see your point. Since your music organization is more personal, less general. Still, you could have a piece of software with virtual folders, instead of duplicating files, that would be neater in my view :]. Then have a button, that opens up the playlist in your favourite music player (like MusicIP Mixer software has).

 

I'm really curious, whether Ross find any potential in our blabber. Would any of these systems (mood, genre tags/virtual folders) be of any use for him?

 

By the way, what languages you speak, Watterson? I'm mostly C# speaker.

[Ross Scott]

Sorry for the slow replies:

 

The best shot for that would be creating a program that allows you to tag any song with any number of user generated tags, and have the user go by each song and each tag manually rating each tag's accuracy to each song it's attached to. This would take a long time to calibrate by the user, especially because I've tried this before and I found myself endlessly calibrating each tag's accuracy and found that I change my mind too frequently. If you have more solid thought processes than me though, then the idea would be for the program to generate a list based off of the tags you searched and order them by most appropriate to least appropriate. The way I did it is I used tags for mood and how the music itself plays, but that came with some problems. A song that's both happy and fast-tempo wouldn't give the exact same feeling as something that's happy and slow-tempo, but it shares the same number of tags as something that's sad and fast-tempo, even when it's closer in mood to the one that's happy and slow-tempo, at least in my experience.

I may be misunderstanding, but I've already tried this, it's not an ideal solution since some songs will get overrepresented. So if you have a song that's electronic AND metal, it appears twice as much on playlists. I ideally want every track to have an equal representation (within its own category). As for custom tags, I really think there's enough within ID3 to work with. I think it would only need 2 different fields max, maybe just one.

 

 

Perhaps this could be what you're looking for Ross...

 

http://betanews.com/2015/06/27/find-similar-audio-files-with-mediapurge/

No, I think that's a completely different thing. That just cleans up badly labelled or organized files. I'm talking about custom criteria for organizing soundtracks so that you can get similar sounding tracks on a playlist.

 

Ross, Watterson, what do you guys think?

Would style (aka genre) + mood + tempo organization work for you?

Yes and no. First, while mood can be very subjective, I think tempo is absolutely crucial towards sorting tracks the way I'm envisioning. Mood or situation for listening could work, but the categories you list is part of the problem. You have things like "energetic" and "poignant" that aren't especially helpful for manual sorting. I want to boil these down into as few categories as possible (but still practical). When you have thousands of potential mood / style combinations like this, it makes manual sorting rather impractical. Besides, where's the benefit of labeling a track "aggressive" AND "angry"? This is way too much. I'm trying to find a system simple enough I can practically follow a flowchart as to how to sort the music and at the end the tracks basically sound similar for that category.

 

Guys, here's the thing: If I know the CRITERIA for how to sort thing, then everything falls into place for me. This is a subjective thing and not everyone will have the same opinions, but the more feedback I get on this, the more I can figure out my own system that works. Lately, I've been trying to put some categories into different moods, and some based on situations where I might want the music. Here are a few examples of what I mean for soundtracks (this all assumes no lyrics):

 

Racing - Rock: Music with a fast enough tempo to be suitable for a racing game with a rock / metal genre overall. Somewhat neutral in tone, without strong "strife" or "angst" overtones.

Example:

 

Frag: Music suitable for a deathmatch or fighting game with pretty constant action. A slower overall tempo than "racing" with more irregularity in the beat also suitable. Generally some variant of heavy metal.

Example:

 

Scary Calm: Music suitable for a horror game, but is generally just mild tension building rather than moods of intense suspense or fear.

Example:

 

Medieval Simming: Music suitable for some sort of medieval sim game. Generally calm pacing, with nothing too soothing or riveting, but has some instrument use or style that suggests a medieval environment.

Example:

 

 

Now the alternative to this would be to have some sort computer-assisted system that could spit out analytics or suggested organizational system that could be refined manually, but I don't know of anything like that. If I have to sort it all myself, the variables need to be as few as possible to still work. I'm anticipating a few dozen categories, but hopefully not much beyond that for soundtracks. Let me know if you have any feedback along these lines. I'd like to know how other people decide what sorts of songs go together (that's NOT based on artist / album).

[Watterson]

I doubt I can truly help here. I merely intuit everything I do with music organization, I don't have a mind that works with flowcharts in this regard. But if somebody could help, likely they'd need a simplified example of what you do in the form of a flowchart.

 

Like, for example, if you use Genre, Purpose of the playlist and BPM to determine the contents of the plalist, an example would be:

 

 

Or something to that effect. From there people of the right mindset might be able to logically add onto/modify the process you use to better suit what you're looking for. I could try myself, but again, I don't really have a mind for that kind of stuff. Sometimes a thousand monkeys and a thousand typewriters can produce results in a reasonable amount of time though, so who knows?

 

EDIT: I also just now realized your most recent post was 2 months ago. I remember seeing it earlier and dwelling on a solution but I only got around to posting this now. This thread not being the most active thread in the world might be a note to temper any expected results with.

 

EDIT2: Also, just a song I feel you might personally enjoy:

https://www.youtube.com/watch?v=zh1RDwUYZlw -- It's a dark folk song by The Pine Box Boys, and the singer sounds a lot like Goofy. You might get a kick out of it if you imagine it's Goofy singing.

Edited November 16, 2015 by Guest (see edit history)

[dad]

Sorry if this answer isn't useful or what you're looking for, but if you want my opinion, it sounds like the best way to sort songs would just be to do it manually, not by any "system." Put songs where you feel they should go; if you later find they don't fit there, move them. With refining and time, it'll eventually get to a sustainable and consistent point. Like you said in the follow-up episode, there's no kind of service like Pandora that will sort music by similar style for video game soundtracks. If you go to Pandora and try to listen to video game soundtracks, it just gives you a jumble of video game soundtracks-- one minute you'll hear boss music from Final Fantasy, and the next it's the overworld theme for Super Mario World. I have a blessing, I suppose, in that I like listening to music not by mood, but by if it's good. I just have a big playlist of music I like, most of it video game soundtracks, some of it not.

[Furcas]

So I was re-watching the Strife video the other day, and I completely forgot about this request. But ever since I picked up Dungeons and Dragons and I wanted soundtracks to play in the background while I take the role of the Dungeon Master, I've been on a similar quest.

 

My method so far has been labelling what sort of emotion I want to convey, rather what I think the music is trying to convey. For example, I have started out with 'Medieval Tavern' for music I would expect to play in a tavern, while 'Mysterious' will be played while my party is in a dungeon and isn't in any danger. 'Sadness' is self-explanatory, as is 'Battle'. 'Epic' is for when you need a bigger sound for a particularly notable moment, such as boss fights or a setpiece. I need to work on the naming conventions and start branching it out into different genres, but if I make any notable progress, I'll update you all.

 

The other thing to note is I'm using Spotify for my playlists. This is less than ideal for Ross's purposes, but I'm hoping I can train the machine learning system that recommends songs to recognize the differences in moods. It's a long shot, but I'm going to try anyway.

 

I'm not sure if the change in perspective is helpful in any way, but I will keep chipping away at this conundrum and see if I can find a system that works.

[Watterson]

First, if the discussion on this has moved locations can anybody provide a link? There hasn't been much Ross interaction here for a while but he still seems very interested in the topic, so he must be discussing it somewhere, just maybe not on the forums?

 

But second, upon watching the Great Berate interview, something clicked with this issue that might help potential solutions -- Sorting songs into categories seems to be less the issue at hand, and playback seems to be what he's after, largely. With this reframe my personal solutions differ and deepen. For instance, third:

 

My new vision of what he's after is a program where you can manually sort sound files by mood/purpose, and tag them with tags like low/medium/high tempo, futuristic synth/retro synth/party synth, lyrics/no lyrics, etc. I'll explain further, but here's a sample video: 

 

 

So you click the umbrella mood/purpose, such as 'relaxing' and the program creates subgroups based on the tags (whose weight might need manual tuning, but likely you'd only need to manually tune a small number of important tags, like weighing tempo higher than the rest), by grouping things that are very close together, leaving files that aren't close enough to an established subgroup adrift between the subgroups, and creating a random shuffle mix, treating the subgroups as a single file in this shuffle, and then shuffling the order of the files in subgroups themselves.

 

With a consistent enough tagging policy, and a large enough sample size, I'm certain there wouldn't need to be much upkeep, so the personal overhead is frontloaded on introduction of the song. Theoretically, at least. In an ideal situation it would also find a way to bridge the gap between these groups, but that would mitigate a lot of the shuffle capability of the sorter. In a case where we want a smoother gradient, it would determine the subgroups, find the best way of organizing the subgroups (as if they were each one whole unit) in a gradient, then put the stragglers (that were sort of close to 2 separate subgroups) in between the subgroups as a mortar, dump the stragglers that don't fit up to a certain standard in this gradient, and shuffle the place of a few of the files each subgroup, careful not to deviate their places too much from the internal gradient of the subgroup.

 

This would require a lot of thought and/or mountains of trial & error though. The returns on investment could very likely not be worth it, and the whole gradient sorting concept might not even work very well, even if it somehow manages to pan out. Also, as mentioned, it would dump some songs, potentially a lot. I guess you could use these songs as interludes between subgroups, like dumping them to their own list and playing one in the middle of the mortar it's most suited to. It would need to be specifically one per bit of mortar though (otherwise it risks compromising the strength of the gradient more heavily) -- which means it could take a long time to get to a specific song if a lot of them belong closest in one specific mortar section. Could take 5 whole loops of the entire shuffle playlist, for instance, if you did it this way (which is the best way I could think of introducing them using this gradient system).

 

My suggestion is to find a program that supports something close to what I outlined, or to seek programming help for putting a system into place like described here and portrayed in the video if it sounds good enough to you. Barring that, honestly, you may want to contact some sort of AI engineer.

Edited December 24, 2019 by Watterson (see edit history)