SongKong Jaikoz

SongKong and Jaikoz Music Tagger Community Forum

Questions about Mood

#9

Great, we have just released SongKong 4.3 so make sure you have the latest version

#10

So I decided to compare results: SongKong 4.4 (Pro) vs. beaTunes 4.6.10.

I picked 10 tracks from 10 different genres.
The results are interesting.

See picture:

Why is there so much difference between the moods, when comparing the outcome of both programs on the same song?

I even dare to say, there are contradictions.
Like for ex. track 5, ‘Extremely Calm’ vs. ‘Very Excited’;
And what about track 10, ‘Very glad’ vs. ‘Very aroused, angry’.

Even the values of Happy/Happiness and Arousal are different.
I thought SongKong was just reading out the data in the database of AcousticBrainz (and beaTunes aswell using its plugin)?

Does SongKong keeps reason with all of the values, like Mood Happy, Arousal, Valence, Agressive, Dance, Party, Relaxed, Sad and divides them to create a distinguished textual tag, like ‘Very glad’ ?
Or is there just a a tag in the AcousticBrainz db written, called for ex. ‘Very glad’?

How does it work?

N.B.: I noticed the picture shows a little vague due tinypic’s compression, so I uploaded the pic, you can download it in higher quality here: http://www68.zippyshare.com/v/aOY5uef5/file.html

#11

Hi, looks like there is a problem there. AcousticBrainz doesnt actually store ‘Very glad’ what we do to work out is use the values that AcousticBrainz does provide to calculate a measure of Arousal and Valence. Arousal and Valence from the two axes of a graph, and when we plot the value that gives us a segment which then maps to the textual values of ‘Glad’, and a vector distance to give us the ‘Very’

The first thing to note is Beatunes Happiness field maps to the Mood Valence value not the Mood Happy value, but there is still a clear disparity between these values as well as the final Mood value.

The problem could be in where they are taking the data or how they are dong the calculation, some songs have more than one set of figures from different user submissions and we may be using a different figure, the fault could lie on my side or with Beatunes.

But what I need from you for these songs is the value of the MB Recording Id column so I can look up the data in AcousticBrainz.

#12

Hereby the MB Recording IDs pulled from Jaikoz: http://www94.zippyshare.com/v/0hvkpEXE/file.html

Track 1: 6941913b-f2c0-4638-a144-a19715df3517
Track 2: cb63543e-c6c9-4233-a3e9-b220101b1699
Track 3: 0519a10c-ad05-40d2-9928-2114472e5d60
Track 4: 0acc282b-c4db-44b2-a6ae-3c037fa3255a
Track 5: 7a6094ca-50e5-4973-bfb0-2c114d574577
Track 6: 5daaf5cf-5a75-43ee-92a2-ee72c6a40a1d
Track 7: 28651acc-1f13-4431-b2de-d93a33a50084
Track 8: 50c1d6f1-f2b2-4925-94c4-f4eacd4bd6f7
Track 9: 5124f0c4-c3eb-49cd-a312-3a4ff43b542b
Track 10: e3ed55e1-fa10-4b30-9cd0-798aef858da6

#13

Thanks Ive checked my figures against some of these songs and they appear to be correct. Im now speaking to Hendrik at Beatunes who is a nice fellow to see if we can work out where the discrepancy between our figures lie, will get back to you as soon as we have an answer.

#14

Spoke to Hendrik and those values are from his last.fm plugin. If you rerun Beatunes using only AcousticBrainz plugin you should get matching results.

#15

Thanks again!

I’ll do some testing asap.

I noticed AcousticBrainz is depending on user input.
I also noticed this blog http://blog.jthink.net/2014/11/how-to-easily-calculate-beats-per.html , where it says:

Is that already the case?

I mean, I do see Create Acoustic fingerprint under Fix Songs -> Match.
But does that mean, I’m contributing to AcousticBrainz aswell?

#16

No it isn’t yet done, it was intended to do it this year but I ended up spending alot more time then anticipated on the Classical music support. But there is a standalone tool available form acousticbrainz.org that you can use to submit data from your own files, then SongKong would be able to make use of that data.

Although similarly named Acoustid and AcousticBrainz are completely different projects, their only connection being they are both MusicBrainz aware.

#17

Thanks, I just installed that AcousticBrainz Extractor and submitted 133 tracks. It skipped 53 songs due lack of info on MusicBrainz-end.

So this means, it found half of them and now the info about Mood is available for the other half of the songs when using SongKong?

#18

It wont happen immediatley because acousticbrainz data is downloaded from acousticbrainz into albunack which is used by SongKong and allows much faster access then if download directly from acousticbrainz

#19

Ah, ok thanks.
Any Estimated Time of Arrival available?

#20

It might be a bit a bit longer than normal what with Xmas and all that, we also plan to have songkong looking up directly form acousticbrainz when no record found on albunack, and that might be in next release.

#21

Testing, attempt 2:
This time using SongKong 4.4 (Pro) vs. AcousticBrainz mood-plugin 0.9.2 from beaTunes with the same songs.

And the winner is…
Amount of songs found and labeled with a mood-description: SongKong
*beaTunes misses 3 out of 10. I tried several different settings, but it wont show any mood for those 3.
I should do such test with my whole collection (read 10k+) to see if there’s indeed a big difference mood-tagged/or not-wise.

The mood-descriptions resemble quite the same this time…
What I do find odd is looking at the result of track #6; Extremely calm vs. Excited, astonished.
Can I look such a song up myself to inspect their values in the AcousticBrainz db mood-wise?

#22

You can look up the numeric values using

http://acousticbrainz.org/mbrecordingid

but it wont show the textual value since this is derived from the numeric values.

Track 6 can be looked up here

http://acousticbrainz.org/50c1d6f1-f2b2-4925-94c4-f4eacd4bd6f7, there are five different submissions with different data so I guess Beatunes is using a different submission.

It may make sense when there are multiple submissions to take the ‘average’, but SongKong doesn’t do this and I don’t think Beatunes does either currently.

#23

Thank you!

[quote=paultaylor]You can look up the numeric values using
http://acousticbrainz.org/mbrecordingid[/quote]
This url doesnt excist though.

I’ve let SongKong run last nite; 10957 songs in total / 1981 tagged with a mood.
I’ve deleted the mood-tags and let beaTunes run today; 7257(!) tagged with a mood.

During xmas, my computer was uploading/extracting my collection with the AcousticBrainz Extractor. It took approx. 35 hours to complete;
10824 in total / 7612 feature files submitted / 3211 files had no MBID and were skipped.

Am I right to say, this is where the difference in amount of mood-tags come from? Whereas beaTunes directly tries to lookup the track in the MusicBrainz and AcousticBrainz database, and SongKong using the indirect approach using Albunack, which isnt updated yet?

#24

[quote=snowseals]Thank you!

[quote=paultaylor]You can look up the numeric values using
http://acousticbrainz.org/mbrecordingid[/quote]
This url doesnt excist though.
[/quote]
Its in italics to signify that you replace mbrecordingid with the actual id as I do below.

[quote]
Am I right to say, this is where the difference in amount of mood-tags come from? Whereas beaTunes directly tries to lookup the track in the MusicBrainz and AcousticBrainz database, and SongKong using the indirect approach using Albunack, which isnt updated yet?[/quote]
Yes, the problem is that MusicBrainz used to provide datadumps on a monthly basis for the new acousticbrainz data. But they have stopped doing that so now I need albunack to trawl for new acousticbrainz data or lookup acousticbrainz directly from SongKong. I’ll probably end up doing both, and it is a top priority.

I think its worth noting that Beatunes would not have found so many moods unless you hadn’t ran FixSongs on SongKong first since that meant SongKong added the mbrecordings ids to the songs making it easy for Beatunes to just lookup the mood using the mbrecordingId rather than having to find a match in the first place.

#25

Noted 8)

Good stuff :smiley:

Correct.
I let beaTunes run over a backup of my collection (11053 songs) - which never been fixed by SongKong - and mood-tagged 4465 songs :!:
So SongKong definitely helps :wink:

#26

SongKong 4.6 can now lookup from acousticbrainz.org directly when no acoustic metadata in the Albunack database. This should give much better percentage but note you still need a match to MusicBrainz, and only some MusicBrainz recordings have so far been analysed for acoustic metdata.

#27

Cool!!!

For the record:

  1. If music doesnt excist in MusicBrainz collection, How do I get my music-collection into MusicBrainz?
    Do I need to manually create an ID like shown here: https://musicbrainz.org/doc/How_to_Add_a_Release ?

  2. If excists in MusicBrainz collection, then they also need Acoustic-fingerprint to match with AcousticBrainz.
    If excists, all good, your program will end-up with mood.

If it doesnt, I upload my collection with https://acousticbrainz.org/download client-software right?

#28

[quote=snowseals]Cool!!!

For the record:

  1. If music doesnt excist in MusicBrainz collection, How do I get my music-collection into MusicBrainz?
    Do I need to manually create an ID like shown here: https://musicbrainz.org/doc/How_to_Add_a_Release ?
    [/quote]
    Yes, if the release already exists in Discogs you can add it more easily using http://www.albunack.net

Correct, I would like to add similar functionaility for submitting data from within SongKong itself but not yet done. Note you should run the acousticbrainz client after matching from SongKong so that the files metadata have MusicBrainz Ids so that the link is created between MusicBrainz/AcousticBrainz. Then run Fix Songs again to update your songs with the acoustic metadata from AcousticBrainz.