Import beat grid feature

I'm currently editing a bunch of tracks in preparation for a wedding and whilst the dynamic track analysis function seems great it's fallen down on some obvious song hits.

Currently Just Dance by David Bowie has been analysed incorrectly (weird given how obvious the drum transients are in the waveform) and I'm having to go through and correct most of the beat grid. It would be nice if there was an import/export beat grid function and a community run database. I'm sure there'd be no copyright issues given that the songs themselves aren't being shared and people could specify which version of the song their beat grid applied to (ie Let's Dance iTunes single release). You could even have a function where the waveform is analysed and preset versions are suggested from within rekordbox.

Alex Blanden 30 mai 2018 12:37 Répondu

Commentaire officiel

In theory, this is a great idea, but in practice, it would possibly be more work to get it functioning properly than it would be to just improve the code to process the track analysis in a more effective manner.

In the case of the CDDB example (which I used a LOT and loved), that matching is done on loose criteria and you are selecting some very basic data to pair with your song. But when it comes to a beat grid, there's much more room for error.

Your example is a perfect display of why this is fraught with potential problems -- David Bowie's song is "Let's Dance," not "Just Dance." Even if you got the title right, the iTunes version you downloaded may not be the same one I did (bit-for-bit) because I've chosen to download the ALAC version whereas you have a 256k variable bitrate MP3. This means that the grid won't align exactly the same due to differences in the headers of the files (and other technical reasons).

Again, a great idea, but I think improved analysis methods are a better place to make efforts for our engineers. ;)

Pulse 11 juillet 2018 17:24

Cette publication n’accepte pas de commentaire.

Date Votes

3 commentaires

Agree the principle of the social (community run) database: just like CDDB used to provide user-contributed info for CDs, remember? If music users and publishers alike could upload beatgrids (and suggested cues) with a option to 'like'/upvote/share just as we do on social media with valued information, that could just save gazillions of deejay man hours painstakingly 'reinventing the wheel'... I can think of better things to do in this weather!!

anotherpiouser 9 juillet 2018 11:08 0 votes

Understood. ID'ing a track would probably require accoustically fingerprinting (Shazam-style). There's open source tech available to do this, such as:

EchoNest's Echoprint - https://github.com/spotify/echoprint-server/blob/master/README.md

GraceNote - https://developer.gracenote.com/sites/default/files/web/html/Content/Music%20Web%20API/Audio%20File%20Fingerprinting.html

Perhaps combining accoustic print with track duration, should give unique track ID that is type-agnostic & header-agnostic. Would also easily pave the way for a libary duplication view feature (hear this requested a lot in forums for RB & other products!)

Waxing lyrical: not saying the ideas kicked around here are the long term solution, but with the future uptake of dynamically obtained streamed content (Beatport/Pulselocker etc.), there will likely be be a correspondingly increased appetite for pre-gridded/cued track metadata, not quite sure how this will converge.

Absolutely we welcome improvements, though it's hard to see how anyone will ever produce an exceptionally reliable analysis, dynamic or otherwise, particularly with tracks having complex/creative rhythm patterns and a vast catalogue of older, non-electronically produced, tracks.

anotherpiouser 12 juillet 2018 14:31 0 votes