Malcolm Ocean's roamblog

optimal streaming-music player & organizer

created 2020-04-30 · edited 2020-04-30 · open in Roam
  • I am so confused given the near-commodity status of the *content* of Google Play Music and Spotify (& Deezer & others) that they aren't innovating more on interface, to lock people in.
  • Obvious things to add:
    • better sharing & collaboration features
    • better listening history
    • better discovery (some of this has actually been *removed* from GPM lately)
    • better track metadata & search (bpm? existence of vocals? much of this can be automatically detected with today's tech)
  • Less-obviously, here's my killer feature
    • There are two types of playlists, I claim. One that is a specific set of tracks in a specific order for a specific context. This is a playlist like "Halloween Party 2009" or "Focus Day Playlist".
    • The other type is a collection of related songs, grouped together because they suit a particular mood, or share some theme, or something like that.
    • (Other possibilities might include)
      • a mixtape
      • my earworms playlist, which is intentionally sorted by "order added"
    • The second type of playlist shouldn't be a playlist at all, but should be a *TAG* that can be applied to a song or a group of songs. This would allow:
      • intersections: playing all songs tagged both #​electronica and #​sad
      • radio stations based on an entire tag not just one song or artist (this already exists for "albums")
        • (oh! it seems GPM has a radio for a playlist...)
      • machine learning to detect which other songs should be part of a given playlist
      • machine learning to generate a playlist that gradually transitions from one tag to another
        • morning: start with smooth jazz, gradually transition to electrofunk
        • evening: start with chillstep, gradually transition to ambient
      • matching your tags with other users' tags, based both on similarity of the tag itself and similarity of the songs
  • If you went a step further and allowed tags to specify whether they were about the lyrics or about the genre, you could use machine learning on the lyrics too and have even more precise radios - songs about a particular topic, or that are particularly full of wordplay or something. and again, this could be combined with genre
    • I'd love to be able to tag a bunch of lyrics and a bunch of sounds, and say "can you find songs out there whose lyrics are like this but the genre is more like that?"
  • I seriously don't know why these companies aren't doing this. it seems like it would simultaneously lock users in more (it's already kinda hard to transfer playlists, but that becomes near-impossible as the options for organizing your music become more complex and unique) and would also create more value in the whole ecosystem because you could explore music based on other users' tags. and ALSO it's like, getting your customers to tag your machine learning training data for free
  • speaking of which, switching from just Like and Dislike to also have the ability to say "this track is fine but doesn't belong on this radio station / this auto-tag". I bet there are people who would be excited to be some of the first to listen to new untagged music and ensure it gets well-tagged.
  • seems like last.fm might be the folks to implement this
  • Notes/links
  • This as tweetstorm - now live here
    • Okay, time to rant about audio streaming services. Why are they all the same old boring things and stuck in the same old metaphors of "playlists" and "radio"?

      Virtual ports of old music-library managers.

      Lots of improvements to be made, particularly: TAGS.

      #ThingsIWantToExist
    • Here's my initial list of obvious things to add/improv:
      - sharing & collaboration features
      - listening history
      - search
      - discovery
      - zoom in / zoom out (preference exploration)
      - custom track metadata
      - artist monetization

      (will elaborate on each of these later in the thread)
    • Here's the killer feature tho: TAGS.

      There are at least 2 types of playlists:
      1⃣ songs in a specific order for a specific context (eg "Halloween Party 2009" or "Focus Day Playlist")
      2⃣ collection of related songs, grouped together by mood, theme, etc (eg "Chill" or "Uplifting")
    • The 2nd type shouldn't be a "playlist" AT ALL, but should be a TAG that can be applied to a song or a group of songs.

      This would allow many powerful new playflows (like "workflows", get it 😉)

      For instance: intersections, ie "play all songs tagged both #electronica & #sad"
    • those intersections could be done based on exclusively songs you've directly tagged yourself, or could use some combo of machine learning (on lyrics & metadata, as well as the audio itself) and other-users-using-same-tags to generate larger source song sets to work with
    • Of course, in addition to intersections, you could also have unions: "play all songs tagged #funk or #horns or #electrofunk"

      You can do this manually by like, shuffling two playlists together, but it's kinda awkward.

      Or subtraction: "#rock − #christian" if atheist friend over
    • Tags plus audio analysis would also allow algos to generate playlists that gradually transition from one tag to another, to sculpt mood

      ☕️ morning: café jazz→electrofunk

      🕺 dancing: "5 Rhythms" [Flowing, Staccato, Chaos, Lyrical, Stillness]

      😴 bedtime: chillstep→ambient
    • If you allowed tags to specific whether they were about the lyrics or the genre, you could use ML on the lyrics too and have even more precise radios:
      - songs about a particular topic
      - songs that are particularly full of wordplay
      - could combine w genre for eg "chill love songs"
    • I seriously don't know why companies like Google Play Music, Deezer, Spotify, etc, aren't doing this. It would increase lock-in and create more value in the whole ecosystem because other users' tags add meaning.

      ALSO, it gets your customers to tag your ML data for free!
    • speaking of which, switch from just 👍/👎 to also have option "this track is fine but doesn't belong on this radio-station/auto-tag, in my opinion"

      I bet there are people who would be excited to be some of the first to listen to new untagged music and ensure it gets some tags.
    • (Though while I'm talking about 👍/👎, I really wish I could tell Google Play Music to NEVER play a song for me. I love @BT's ESCM album except the track Solar Plexus, which I have to remove/skip manually ever time I put the album on 🤦‍♀️…awful disempowered feeling.)
    • I want to elaborate on some of the things I described in the second tweet.

      I think why these "obvious things" haven't happened yet, is that the creators of these *apps* haven't escaped the old metaphors of "music library" & "radio", even tho new bizmodel.
      twitter.com/Malcolm_Ocean/status/1268929904680321030
    • Better sharing & collaboration features.

      Little innovation on this since I first tried Rdio party playlists in 2012, & GPlayMusic is still worse than that.

      Ideas:
      - one curator, many suggesters
      - built in "name that song"
      - user-contests for recommendations
    • Listening History. Maybe other apps do better than GPlay here.

      But this could be really powerful, with audio & graph search features like "I heard it last week… at least 5 mins long, and had a kind of slow opening, and it was just before some Coldplay track"
    • Speaking of Search, why the FUCK can't I create playlists filtered by bpm? (ie beats-per-minute, aka tempo)

      It's super easy to calculate bpm of a song (for most genres) and this would be awesome for making running playlists or restricting chill playlists from getting too upbeat.
    • Or either in the form of search or as a radio filter, being able to force instrumental tracks only, or only ones with vocals, or no drums at all, or minimal heavy-bass (I mean, having built in equalizers would help with that, but still)

      or to search/filter by year/month released
    • Speaking of year/month released, better Discovery.

      GPM nuked theirs last year (I think tryna push ppl to YT Music) but it was also only like a 3/10 before, by my reckoning. Soundcloud is somehow like a 7 or 8/10.

      10/10 would have a mix of precise tuning & feedback + magic
    • I think these services maybe focus on being able to provide you with an endless stream of music that you'll find "decent"—which is an amazing resource and not to be knocked, but!

      ADDITIONALLY we need modes where half are total meh and 5% are new fav songs/artists 🤩
    • Relatedly, the ability while listening to a radio or other discovery source, to say "this song is awesome, can you try more of this album/artist on me and see if I like it?" and it would sort of organically do that (in the current radio & later) and find out if you're into it
    • And, corollarily, to say "I really don't like this" and have it sort of zoom out and play less of that artist. Ideally even take direct feedback like "dumb lyrics" or "too harsh noises" or "too repetitive".

      (& this could be tagged with "never" vs "not in the mood atm")
    • Finally, track Metadata. Stick text on songs, both for later reference & search, and for social reasons.

      Could even do playful things like "show this comment to any friend who listens to this track 10+ times" and there'd be an "omg you love this too!" moment
    • Wait, one more: artist monetization!

      People want to support their fav artists, and let's face it: streaming services are shit for this. But they could help, without hurting their own bottom lines, by supporting ticket & merch sales, or… integrating artist patreon perks?
    • Are all of these good ideas? Maybe not. It would certainly be a lot 😂

      But it's weird how few new ideas (good OR bad) we've seen from these companies offering music streaming services.

      twitter.com/patio11/status/1274576125394513921
    • If you enjoyed this design thread, you may like my many other threads of #ThingsIWantToExist!

      Topics include e-readers, conversation-apps, podcast apps, crowdfunding, &+

      twitter.com/Malcolm_Ocean/status/1151673122879352839

6 linked references

February 9th, 2021

tuning

design metaphor of being able to specify something like a… › context-dependent › there should be a much richer way to describe you context…
  • M > relates to my concept of tags in optimal streaming-music player & organizer
    • QC > like how spotify serves you a song and you like/dislike
      • what if it, or twitter, could let you say "this is not the thing I'm in the mood for"
      • QC > tiktok I think is good at this?

me as project advisor / biz consultant / design jammer / co-visioner

projects/people I've done any amount of this with:

January 27th, 2021

Design Docs

April 30th, 2020

I wrote a script to convert WorkFlowy JSON into roam JSON…