|
Mixed In Key (pg. 3)
|
View this Thread in Original format
| Stu Cox |
[Thought I'd drag this thread up again rather than start a new one on the same topic]
Decided to give MIK a go because Rapid Evolution's never got more than about half of the keys right in the tracks I throw at it...
And it's just as bad. Just gave it a stack of 36 tracks and it only got 17 of them right. Which is, to be honest, unusable. In fact I'd say anything less than 95% would be pretty useless: marking a track with the wrong key is worse than not marking it at all.
There were a couple of instances where I would have said a track was say G minor and MIK thought it was D minor, which I guess is understandable and technically might be indistinguishable given limited harmonic content, but if the track strums away on a G chord for the first 2 minutes then it doesn't really make sense to label it as D minor for mixing purposes!
Others were just plain wrong: there were loads of notes in the track which don't even appear in the scale of the key MIK chose, including the note I would have said was the root note. I can't see how the key chosen could be considered to be valid.
I appreciate the processing involved in detecting a key is very complex and can't be absolutely perfect. I also appreciate that a lot of modern dance music might be best described using a large number of accidentals which must be very hard for software to pick apart. But MIK are charging for this product and nowhere on their website does it suggest that it's anything less than 100% accurate. I also appreciate that the key of a track isn't an absolute thing: some tracks could be considered to be in a number of different keys, but really this could only account for a couple of the tracks for which I disagreed with MIK.
Has anyone else had as little success as this?
And yes, I know doing it by ear will always be more accurate, which is what I've been doing for the last 3 or 4 years, but it takes ages so I was hoping MIK might at least come close! |
|
|
| orTofønChiLd |
| i never knew mixed in key had so many errors, wow |
|
|
| ZeJayMan |
| how software like this works (or doesn't apparently) is still witchcraft to me. |
|
|
| Rodri Santos |
i always struggled about how the software detects key and bpm, i find Mixmeister more accurate, i use this software for mashups and i think is more accurate, always detect the tracks as minor key but i think is more accurate, only detected Marco V - Contour (Amo & Navas Remix" as a Major Key but at least seem to detect properly when it's a # or a root , i use this to label my keys with this but put a "?" sometimes because i test it with some tracks i now the key and they don't sound the same.
However for mixing purposes you most notably need bass at the end and at the beginning of the tracks, this is easy to check with your ears. |
|
|
| Stu Cox |
| quote: | Originally posted by orTofønChiLd
i never knew mixed in key had so many errors, wow |
After testing it a bit more today, I should probably withdraw that heavy slating, at least a in part.
I just ran another 20 through it and it got all bar about 4 right this time, 2 of which are perfectly excusable (in fact the key it chose might be more 'correct', but for mixing purposes it made more sense to label it differently) and the other 2 were just plain wrong.
I also went over the keys it chose for the load I gave it yesterday and on reflection I can see how a few more of the ones I disagreed with it on could be valid - and spotted one where I'd got it wrong myself but MIK got it right... oops ;) And maybe it was just a tough bunch anyway - quite a lot of techno in there.
So I think I will carry on using it, but I'll have to check the results each time. This may sound like it negates the point in using MIK in the first place, but if it suggests a key it's much quicker for me to check if that's right or wrong than work it out myself from scratch, particularly guessing major / minor correctly... then hopefully I'll only have a handful to work out myself.
I'd guess MIK's probably 80-90% 'accurate', but personally I'll probably chose to correct 20-30% of the keys it picks (the extra ones being where it's more practical marking it with a different key). |
|
|
| orTofønChiLd |
| quote: | Originally posted by Stu Cox
After testing it a bit more today, I should probably withdraw that heavy slating, at least a in part.
I just ran another 20 through it and it got all bar about 4 right this time, 2 of which are perfectly excusable (in fact the key it chose might be more 'correct', but for mixing purposes it made more sense to label it differently) and the other 2 were just plain wrong.
I also went over the keys it chose for the load I gave it yesterday and on reflection I can see how a few more of the ones I disagreed with it on could be valid - and spotted one where I'd got it wrong myself but MIK got it right... oops ;) And maybe it was just a tough bunch anyway - quite a lot of techno in there.
So I think I will carry on using it, but I'll have to check the results each time. This may sound like it negates the point in using MIK in the first place, but if it suggests a key it's much quicker for me to check if that's right or wrong than work it out myself from scratch, particularly guessing major / minor correctly... then hopefully I'll only have a handful to work out myself.
I'd guess MIK's probably 80-90% 'accurate', but personally I'll probably chose to correct 20-30% of the keys it picks (the extra ones being where it's more practical marking it with a different key). |
wow, when i bought mixed in key, i thought i was mixing 100% harmonically but you say 4 right outta 20, thats a shame. This should be taken off the shelves. |
|
|
| Teezdalien |
| quote: | Originally posted by orTofønChiLd
wow, when i bought mixed in key, i thought i was mixing 100% harmonically but you say 4 right outta 20, thats a shame. This should be taken off the shelves. |
He said all bar 4 out of 20 were right. |
|
|
| Mad for Brad |
all these softwares do is analyse the notes , assumes you are use one key, assume you are using tonal harmony with no accidentals or applied dominant harmonies then once it gets rid of the scales that don't fit the algorythm, it picks the note with the highest number of incidence and voila. They don't really work because the method they use to find a key is flawed. For example, you could be in the key of F minor and never get an f minor chord. But this software would not know what to do and give you its best guess.
IF they had some help on the theory side, i'm sure they could find something that worked on EDM 99.9% of the time even with chord changes and what not. |
|
|
| orTofønChiLd |
| quote: | Originally posted by Teezdalien
He said all bar 4 out of 20 were right. |
ooh lol damn |
|
|
| Rodri Santos |
| quote: | Originally posted by Mad for Brad
all these softwares do is analyse the notes , assumes you are use one key, assume you are using tonal harmony with no accidentals or applied dominant harmonies then once it gets rid of the scales that don't fit the algorythm, it picks the note with the highest number of incidence and voila. They don't really work because the method they use to find a key is flawed. For example, you could be in the key of F minor and never get an f minor chord. But this software would not know what to do and give you its best guess.
IF they had some help on the theory side, i'm sure they could find something that worked on EDM 99.9% of the time even with chord changes and what not. |
the fact that there are updated versions of MIK makes me think that they improve the alghoritm a bit, some versions maybe are also code fixes but i think this software is meant for EDM songs, they're either crappy programmers or machines aren't yet accurate enough.
With the bpms i usually know when the software is going to fail, traktor is pretty bad detecting bpms but has a lot of problems when the 4/4 is synchopated, there are several layers of rhytm patterns ouf of the 4/4 signature like in DnB or Dubstep with 3/4 or 6/8...
I will encourage you to use mixmeister again, the bpm is to the date hardly ever/never wrong. This must be a good software. |
|
|
| kadomony |
i find MixMeister works best for me.
i can just drag my whole music collection into it, let it run and sort the columns by key(code) and/or bpm. plus, it saves the collection.
makes it much easier to locate possible candidates for next tracks when arranging a set in ableton. (also helps when searching for a track when you only know some of the ID tag info)
it is off a little at times, but nowhere near the rates i've seen with other software. |
|
|
| skip |
I really do wonder how these programs detect an entirely wrong key at times. I do not understand much of music theory, but analyzing the frequencies of a track shouldn't be that hard. I don't really know what exactly determines what key a track is in, but I would think that it's pretty easily defined. Probably more focused on the mid and highish frequencies? Probably the amplitudes of the frequencies also play a part?
Anyway, with all this knowledge I can't see why it really is that hard to write an algorithm which isn't more accurate. I know qualia (if I remember his nick right) who develops Rapid Evolution has posted here, so maybe he could shed some light into the matter. I'd be interested in hearing more about it at least. |
|
|
|
|
