Comment by notheguyouthink | Hacker News Reader

Comment by notheguyouthink | Hacker News Reader

Parent Full thread

notheguyouthink

> The answer is that "cool" and "happy" tags won't find the music you want. Vivaldi's "Spring" and Bobby McFerrin's "Don't Worry Be Happy" and "C'Mon Get Happy" by the Partridge Family are all "happy." As is "Sailing" by Christopher Cross and Charlie Pride's "Kiss an Angel Good Morning."

Are those all the same artist? The same album? The same genre?

My point in the above questions is that of course a single field won't perfectly fit to all songs in existence, but "fit" is subjective. If i want "up tempo" music, and i don't request anything else, then give me all up tempo music. Rage rock, rap, symphony, all of it. I can refine the filter if i want.

This allows me to query "play me all up tempo music from Dave Matthews Band". "Play me all happy music from Dave Matthews Band's album Live at the Rocks", and etc.

> Turns out that what most people really want when they say "I want to be able to sort/listen to my music by tags" is something that feels like a curated playlist, where the songs naturally work together one-after-another without being jarring, usually within some music category if not within a narrow genre.

Clearly that's not what i want, as i am extremely dissatisfied with Playlists.

Expecting all music to somehow be grouped into non-flexible folders (which is all playlists are) seems silly to me. Do you think i shouldn't be able to Search Google/Internet for custom terms?

Why can i not search music for custom tags, but i can search for genres? Having tags doesn't mean Playlists can't exist, but it can allow me to refine my UX to exactly what i want. I'm sick of curated Playlists, and i'm tired of trying to remember all song names in existence to build my own. I want better data to build collections of songs.

I don't disagree with you that a well-tagged music repository would be useful - indeed this is the goal of the music genome project - but am trying to point out that the typical requirement for "tagging" music is well beyond almost any individual user. Songs need to be tagged by genre, subgenre, style, mood, tempo, instrumentation, vocalist style, intensity, scale, key, and many more categories in order to adequately be able to mine the database and return results that the typical user would find useful. Then the user would have to be sophisticated enough to form queries that would return useful results. Maybe "uptempo" would be a meaningful result for you but a cross-section of all uptempo music would return unacceptable results for pretty much everyone else.

Who wants to follow Van Halen's "You Really Got Me" with the Endaro Mahaanubhaavulu? Answer: nobody. So you need to define genre. OK, you select "Rock". Now "You Really Got Me" is followed by "Non E Per Sempre." Oops. Still the wrong genre - and also the wrong language.

So you specify "English" words and try to tighten the genre. Now this is harder: is Van Halen "hard rock" or "metal" or "glam rock" (answer: could be any of the above). Pick the wrong one, and you'll accidentally expunge "You Really Got Me" from your listening experience, or include a bunch of Poison and Tesla songs, and you really hate those bands because you think their lyrics are dumb.

Uh-oh. Now you need a way to categorize the thoughtfulness of the lyrics.

The fact is that the requirement that most listeners have is something much more intelligent than a database search. If you've attempted to use the music genome project to find music, you'll know what a difficult process it can be. The music it returns is extremely hit-or-miss - and it is exquisitely tagged. By the way, I do think that the music genome project will solve your particular requirement "show uptempo music from Dave Matthews Band."

I'm not saying it's not possible. I'm saying it requires next-gen AI to (A) adequately categorize music on all needed dimensions and (B) assist the user in forming meaningful queries.

I'm also saying that the experience of radio broadcasting over the last 50 years is that listeners almost always want something that feels extremely familiar and highly curated....

> Who wants to follow Van Halen's "You Really Got Me" with the Endaro Mahaanubhaavulu? Answer: nobody. So you need to define genre. OK, you select "Rock". Now "You Really Got Me" is followed by "Non E Per Sempre." Oops. Still the wrong genre - and also the wrong language.

Of course, that's why you combine filters. You keep speaking of single queries.

    tag:uptempo tag:epic instruments:guitar genre:electronic

Sure, people may disagree with what is uptempo or not, but they already do that with genre's, i know i do haha.

I'm not saying it will be perfect. I'm saying right now we have nearly no useful ability to discover music. I'm extremely dissatisfied with music discovery, and i have no control in most applications.

> I'm also saying that the experience of radio broadcasting over the last 50 years is that listeners almost always want something that feels extremely familiar and highly curated....

Which sucks, because it's clear most popular products (Spotify/etc) agree with you - and i detest it. I have been given zero, completely zero meaningful features to interact with. At best i have magic in the backend to "hopefully maybe implement what i might want". There is no interaction to allow me to try and shape this with any meaningful feedback over what each interaction does.

What does this Heart do? Is it permanent or is it just temporary? How do i build a dynamic playlist off of a mood? How do i remove certain artists/genres/instruments/vocals from the mood?

Most music playing services give me absolutely nothing. And it's terrible, to this user at least. I just want features, and no one is implementing user facing features.

edit: To be clear, i'd settle for community tagging. Hell, i'd setting for the ability to tag posts myself, with no other contribution. I can't even play music by a tag i define. I promise you, i can manage to tag some music myself - but i have no ability to do that. The best i can do is build a playlist, which is like 1990's Mail folders. Completely useless for dynamic applications.

So what about the Music Genome project?

http://www.pandora.com/about/mgp

Reply on news.ycombinator.com