[Hot Topic] Key Detection Software Comparison

Camelot_Sound

Three different software programs now offer key detection abilities:
Mixmeister, Rapid Evolution, and MixedInKey.

Using a highly qualified musician as our reference source, we have found that Mixmeister (MMP5) provides roughly 60% accuracy overall. We have not tested the accuracy of the other programs. I encourage you to compare the accuracy of all three programs, along with their ease of use. For this purpose, here is the latest Camelot subscriber update. It mostly contains information on tracks contained in various April 2006 editions of different Promo Only Club and Radio series. The field with an asterisk reflects our musician' report, followed by MMP5 results.

For example, Trace Adkins' "Honky Tonk Badonkadonk" is reported as keycode 9A by both our musician and MMP5. Both MixedInKey and Rapid Evolution SHOULD also report keycode 9A (E-Minor). MMP5, on the other hand, reported Bigworld's "Me Gusta" as a 6A (G-Minor) while our musician reported it as 8B (C-Major).

DJ Andre

Why do you think MixMeister company decided to only read MINOR keys than also include MAJOR keys?

Camelot_Sound

I'm really not sure. I vaguely recall an explanation on the MixMeister website, but I didn't find it in a cursory glance through posts revealed by using the forum's FIND function for "major keys." There's a lot of posts there by DJ Andre, however . . .

DJ Andre

Camelot_Sound

Yes, the vast majority of dance music has been in minor keys for the past twenty years. Rock, country, and other genres that are normally performed by bands, rather than being produced by computers, continue to use major keys. Many ballads and sample-based songs, such as Paul Wall's "Girl," are also in major keys.

It's interesting to note, however, that in recent years major keys seem to be making a comeback in dance music and R&B/Hip Hop.

coke

Camelot_Sound

It's my understanding that when MixMeister reports a major key, it is actually a glitch in the program instead of a valid report.

DJ Andre

coke

DJ Andre

I got a chance to contact Oslo about this.. he said he was gonna investigate and run a new query to see what he finds out on that Major key.

qualia

hey everybody, i'd love to do this comparison as well. the problem for me is getting a hold of copies of these songs (legally). i realize they're taken from the top charts, but i guess i don't have a very popular taste in music lol ;-) if anybody has any ideas on how i can get these songs, i'd gladly run rapid evolution and possibly these others apps and compare.

i thought i'd also mention, it would be curious to add the mixshare.com results as an item for comparison. in case you didn't know, it's an online database of musical key information (rapid evolution automatically integrates with it). it represents the net result of over 300,000 user submitted songs! i'd be interested to see if the results shared here are any more or less accurate than automatic detection.

i must say that the mixshare server is still in early stages, and a lot of cleaning up of the data needs to be done, but it's designed to aggregate multiple results for the same songs to determine the most probably bpm/key from what people enter. theoretically, the more people the use it the more accurate it should get. it could even eliminate the need for automatic detection...

coke

about mayor keys:

you can go to preference in a song a manualy change the key to major... that can be the reason of the majro key on the mixmeister results bandhead

Camelot_Sound

You can probably order Promo Only CDs directly from www.promoonly.com.

qualia

excellent, am in the process of obtaining the songs. will post results when i have a significant amount of info...

qualia

okay, i only have 64 of the songs so far, but thought i'd share my current results. i will post again only when i have the entire list of songs tested ;-)

total songs: 64

MIXMEISTER 5:
32 correct (50%)
9 compatible (64% correct or compatible)
23 incorrect (36%)

RAPID EVOLUTION 2.2.10:
34 correct (53%)
15 compatible (77% correct or compatible)
15 incorrect (23%)

so it looks like rapid evolution is in the lead at the moment, and when it guesses incorrectly, it has a higher chance of predicting a compatible key which is better than being completely incorrect...

rather than take up space here, you can view the details of my tests at: http://www.mixshare.com/cgi-bin/yabb/YaBB.pl?board=re_board;action=display;num=1147641315

as time permits, i also plan to see if using the live audio input key detection of rapid evolution over the Chorus of these songs yields any better results...

Camelot_Sound

Great start!

When comparing results, it's important to use the same version of songs. I learned the hard way that software may report different keys even when the versions are actually in the same key. I can only speculate that the intro makes a difference.

For example, after using SoundForge to extend a song, I occasionally find that MixMeister reports my extended version as being in a different key than the original version from which it came. This reinforces the value of using the same song version when comparing software. For this reason I recommend using the Promo Only CD versions of these test songs rather than MP3 versions downloaded from other sources. I understand that the trial version of MixedInKey handles wav files but not MP3 files.

If your test results for the same song version differ from our results, then we may safely presume that differences are software based.

qualia

hi camelot, was reviewing the keys rapid evolution got incorrectly, and saw 2 so far that i think your info might be incorrect for (i put down what i think is the right key):

b.g. ft mannie fresh - move around 6A
goldfrapp - ooh la la (stent version) 8A

can anybody else check these and confirm? i'm not sure how much else of camelot's info might be incorrect, but we should probably try to have knowledgable people confirm these results too...

qualia

as far as results varying from different versions of the same song... once i have all 100 songs here, i will run mm5 (since i own a copy) and mixedinkey (if djyakov will send me a copy, hehe) against the same exact songs that i ran rapid evolution against and compare... should eliminate any of these differences :-)

Camelot_Sound

Good job, Qualia! Yes, we occasionally make mistakes so I will have our musician confirm those two keys. The difference in "Move Around" is between G-Minor (6A) and G-Flat Minor (11A). I would not be surprised if the true key is between these two.

I made a typo in reporting Ooh La La (Stent). Our musician actually reported this version as being in A-Major (11B) rather than A-Flat Major (4B), as I posted. It should therefore be tested to verify A-Major, or A-Minor as reported by MixMeister and Qualia's RE.

qualia

yeah, the goldfrapp song is 11B (A major). i was working quickly through the list and didn't spend time to see if it was A major or A minor, i just heard that the root note was definitely A and assumed RE and MM5 both got it right ;-)