I own a very large collection of music of all genre (classical, jazz, pop, rock, electronics, experimental) with > 100.000 songs and I was frustrated when I could not find a very interesting song I ripped some time ago but forgot both its name and artist. So I searched for a classification (tagging) scheme I could use on any song of any genre which goes beyond the usual playlist type of tags - which do not work for very large collections because the resulting lists of matches are still way too comprehensive.
I could not find any such scheme even after many hours of research so I took the effort to develop one by myself. Meanwhile I applied this to several thousands of songs of different genre and I refined it until I was (somewhat) satisfied with the search results. The aim was to have a system where 3-5 different tags of the most identifying song characteristics lead to matching lists which can be used both as a mood-based playlist AND a specific song search list with a reasonable (limited) size.
I defined 20 characteristics with 6 values each, mostly alternative (numbered 0-6) and some multiple-choice values (<attribute>). A starting value numbered 0 includes the blank value for that characteristic too (e.g., "Power: 0-upstage" includes "no power at all" and "distant power") while a starting value 1 does not include the blank value (e.g., "Energy: 1-crank up" does not include "no energy at all"). Most characteristics DO NOT include very common hence not distinctive attributes (e.g., standard sound or adaptation, happy lyrics, sentimental mood). Why tag ordinary stuff? And I did not use any words which are included in genre or style categories (e.g., dark, ambient, martial, hard, psychedelic) to avoid confusion.
So here the system is:
TEMPO: 0- grave 1-adagio 2-moderato 3-allegro 4-presto <alternating>
DANCE: 1-foxtrot 2-disco 3-rave 4-rock&roll 5-old school 6-unique
RHYTHM: <groovy> <boppy> <infectious> <bluesy> <swinging> <jazzy>
PATTERN: <marching> <rocky> <funky> <boogie> <waltzing> <latin>
BEAT: 1-monotonous 2-lively 3-stamping 4-driving 5-nervous 6-erratic
ENERGY: 1-crank up 2-tap foot 3-hum along 4-sway along 5-sing along 6-headbang
SOUND: 1-vintage 2-trippy 3-fantasy 4-mysterious 5-futuristic 6-cacophonous
POWER: 0-upstage 1-vibrant 2-forceful 3-wild 4-brute 5-stoned
ADAPTATION: 1-austere 2-airy 3-thrilling 4-complex 5-mellow 6-bombastic
MOOD: 1-upbeat 2-laid-back 3-melancholic 4-gloomy 5-scary 6-aggressive
SENTIMENT: 1-cheesy 2-romantic 3-touching 4-solemn 5-uplifting <spiritual>
TEMPER: 0-clinical 1-tense 2-threatening 3-ominous 4-apocalyptic 5-hostile
TUNE: 1-dissonant 2-peculiar 3-harmonious 4-magical 5-catchy 6-earworm
MELODICS: 1-dreamy 2-smooth 3-cheerful 4-captivating 5-cerebral <concise>
VIBES: 1-funny 2-patriotic 3-shamanic 4-demonic 5-ethereal 6-sensual
VOCALS: 1-sad 2-soulful 3-cherubic 4-jolly 5-roaring 6-harsh
VOICE: <legit> <distinctive> <expressive> <sombre> <falsetto> <recitative>
LYRICS: <silly> <ironic> <juicy> <queer> <vulgar> <dadaistic> <intellectual> <political> <carol>
ODDITY: 1-strange 2-trashy 3-quirky 4-weird 5-freaky <extreme>
Xtra: <epic> <eclectic> <theatrical> <solo> <riff> <cover version>
I implemented these values in the JRiver field [keywords] in the format "Dance: 1-foxtrot" where they appear in a drop down list for easy tagging and - most importantly - they are stored in the file tags so they do not get lost when copying the files to a different destination. Usually I pick 3-6, up to 10 (if I like the song), but only for songs I want to remember and listen to again for whatever reason (again: Why tag songs that do not matter?). In combination with the song rating, genre-style and/or personal playlists this scheme works pretty well so far - but it is a lot of individual tagging work to do, no doubt. So this is for music freaks only, I am afraid.
Please feel free to use this scheme or parts thereof if you find it appropriate. Any comments and suggestions are highly welcome.
Author
Topic: Music (song) classification system for very large collections of all genre (Read 2246 times)