Reanalyze tracks in 5.2.3?

Richard's Avatar


19 Dec, 2019 06:32 AM

Hello Hendrik,

I see in the note that v 5.2.3 made some changes to the tempo and key analysis. Do you recommend that users re-analyze their tracks on the key and BPM items?


  1. Support Staff 1 Posted by hendrik on 19 Dec, 2019 10:22 AM

    hendrik's Avatar

    Hey Richard,

    tempo and key estimation accuracy of v5.2.3 should be a bit better than in previous versions. Whether you want to re-run analysis is entirely up to you. If you just have a few thousand tracks or accuracy is super-important to you, I say go for it and let it run.

    For tempo, the increased accuracy will manifest in fewer so-called octave errors, i.e. error by a factor of 2 or 1/2.

    Happy holidays,


  2. 2 Posted by Richard on 20 Dec, 2019 06:51 AM

    Richard's Avatar

    I've noticed in the BPM column of the library listing, sometimes the 2nd (rightmost) BPM number is in a dimmed font. Other times it is "standard" weight. Does that signify anything? Just curious.
    Also, I spot checked a few tracks and the new 5.2.3 tempo and key calculations are in some cases producing different results than before. Sometimes the actual tempo changes by just a small fraction, sometimes it results in an octave change, just as you said. Also, I noticed some key changes. So I decided to reanalyze everything. I do like accuracy. One thing of interest: It appears to be better at actually generating the metadata per track. 80,000 tracks in and the Task Queue has yet to generate an "I can't figure this out" message for any track. In the older version, every now and then it threw an error on a "random" track. (It probably wasn't actually random, but I didn't investigate further.) 12 hours in, two days to go.

        On Thursday, December 19, 2019, 4:22:05 AM CST, hendrik <[email blocked]> wrote:


  3. Support Staff 3 Posted by hendrik on 20 Dec, 2019 09:58 AM

    hendrik's Avatar

    Hey Richard,

    the tempo algorithm always produces two values and a so-called salience value which is inspired by the P-score metric used in the MIREX automatic tempo estimation task. The salience value indicates which of the two values seems more likely (it actually attempts to model tempo perception—different people perceive tempo differently). In the UI this is reflected by the second value and letting it appear either gray or black depending on how likely it is that someone would actually perceive the track in that tempo. You can see the perceptual distribution/certainty by hovering over the value with the mouse. The tooltip shows the actual number.



Reply to this discussion

Internal reply

Formatting help / Preview (switch to plain text) No formatting (switch to Markdown)

Attaching KB article:


Attached Files

You can attach files up to 10MB

If you don't have an account yet, we need to confirm you're human and not a machine trying to post spam.

Keyboard shortcuts


? Show this help
ESC Blurs the current field

Comment Form

r Focus the comment reply box
^ + ↩ Submit the comment

You can use Command ⌘ instead of Control ^ on Mac