I have a mp3 file that in rekordbox has the begin of the first beat at 0.519. Thus the beat is audible from 0.519, the grid startpoint is at 0.519 and the waveform clearly shows the beat starts at 0.519.
However when this is exported to USB and played back on a XDJ-1000mk2 the audio of the beat starts at 0.495, while the grid startpoint is at 0.519 (as analyzed by rekordbox).
This causes the grid and the audio to be off by ~24 milliseconds, and thus causes problems with sync, phase meter, quantized loops and cue points.
The audio file: http://www.bvwebdesign.nl/qbDV32ejbHhRaKnv/dakada.mp3
(free download from: https://soundcloud.com/guerillacrewrec/shiso-dakada )
rekordbox: 4.3.1 osx 10.11.6
XDJ-1000mk2 firmware: 1.22
Publicação fechada para comentários.
@Jan > Can you try deleting the track from your USB device, re-analysing within rekordbox and then export and try again? I want to make sure this is, in fact, an issue with the XDJ-1000MK2 (which I don't have on hand to test with, but every other device I have loaded properly with the beat @ 0:00.519).
If you can confirm that after testing, I can escalate this to our product team.
Yes, the issue persists after re-analyzing. Furthermore, I have also tested it by just putting the mp3 directly on an other USB drive without using rekordbox, the results are the same.
Are you absolutely sure the audio of the beat starts at 0.519 on other Pioneer players? I do ask because the difference is rather small at only 24 milliseconds and at first glance it seems the beat does start at 0.519.
The method I used to determine the difference is by setting the player in reverse mode. When starting from a cue point at 0.519 it should be silent, however it is not, you can hear a part from the beat. Only when starting in reverse from 0.494 there is a complete silence.
It was bang-on at 0:00.519 for every player I tested it with (CDJ-2000NXS2, CDJ-2000NXS, CDJ-900NXS, XDJ-1000, XDJ-700, XDJ-RX), and the audio matched the visual representation of the waveform.
I'll pass this up to the XDJ team and see if they can test it to provide any other info.
So here's what the engineers had to say...
The mp3 you provided is encoded by Lavc57.24, and it includes an Info tag(XING tag) in the MPEG frame. The CDJ cannot handle this information normally, as this is different from the mp3 standard file which is encoded by popular LAME encoder.
They are investigating ways to handle these issues but can't guarantee a solution.
My recommendation would be to first try removing that tag, and if it doesn't resolve the frame offset, to cross-code with a different encoder. That's not preferable as it would reduce the quality, but at least it would clear the error.
Thank you for the information. I was able to remove the Xing tag with ffmpeg (this command copies the original audio data, artwork and all tags except the Xing one):
ffmpeg -i dakada.mp3 -write_xing 0 -c:v copy -c:a copy dakada-without-xing.mp3
This does the trick as the new file has the same audio offset in rekordbox as on the XDJ.
Funny thing though, the first beat is now analyzed to be at 0.493 in rekordbox. Same as the XDJ recognized it in the first place (well a 2ms difference that can be contributed to measuring error in the first place). Suggesting that the bug is in rekordbox (or that ffmpeg and XDJ handle the files in a similar fashion).
@Jan > Gotta love the power of the command line! I know such functions exist, but I rarely recommend them as I've seen users brutalize their files simply by missing one small portion of the string. ;)
The engineers weren't certain where in the decoding it occurred, but I'm glad you were able to resolve it and thanks for providing the feedback as to what worked for you!