Calculating MIDI latency?

[Will Russell]

Hi all,

I could use some help from you ProTools MIDI experts.
5,1 12 core Mac Pro
Mac OS 10.14 Mojave
Pro Tools Ultimate HDX 2021.12

I did a live tracking session today with a real guitar, vocals (to a click) and someone playing a MIDI keyboard through MOTU Fasttrack USB, to an instrument channel with the Ivory Grand Pianos II plug-in. 24/96 session.

Headphone monitoring was via sends from each input to analog outputs to a Aviom headphone system.

Although I usually drop down the playback buffer as low as I can for MIDI tracking (128@96k) I inadvertently left the playback buffer at 512. The keyboard player did not report any latency however on playback all of the keyboard tracks are ahead of the established tempo. Apparently the player compensated for the latency by playing everything a little bit early.

Sigh.

It seems to me that there must be some way to calculate the MIDI latency but I can't for the life of me find any information leading me to that information. If I could calculate this latency then I could simply slip the MIDI tracks back by that amount and put them in proper time.


Help! I don't do a lot of work with virtual instruments and any advice would be wildly appreciated!

[albee1952]

It won't do any good to calculate latency when the player tried to compensate. I would fine a spot where the guitar(or choose any instrument) and the piano SHOULD be playing tightly together and slide the midi data to align with the chosen "guide" instrument. Alternately, commit the midi data to audio and slide the audio in the same fashion. Since its reasonable to expect any 2 players to be slightly off at any given moment, I suggest you accept what sounds good and not try for "perfection" on a technical(ie, grid-related) level.

Alternately, you could try to use Beat Detective to extract a groove template from another track and force that onto the midi No matter which route you decide on, I would duplicate the playlist(whether midi or audio) first so you always have the original to go back to, just in case

[JFreak]

MIDI is said to be sample accurate, but in reality MIDI 1.0 spec from 1981 says that MIDI clock is 31.25kbit/s and word length is 8bits.

It means that on a 44.1kHz session there are 352800 bits per second, or 344.53kbit/s worth of MIDI data available. Now, 344.53/31.25 = 11 samples of accuracy.

When you press a key, your chances are you have latency of 0-11 samples and nothing in the world can compensate that. Until we have MIDI 2.0 implemented.

[Will Russell]

Thanks to everyone who has offered help so far!

512 samples at 96k is only 5ms and the piano tracks in question are off by way more than that...

I totally get that in the end it's gonna be a nudge fest based on how it sounds and feels. I was just trying to establish whatever the measured latency was as a starting point. Based on feedback so far it sounds like that's not really gonna happen.

I just realized that another important detail is that I had Sonarworks Reference 4 on the master which I think affects overall latency even though the headphone mix was derived from a send and wasn't going through the master? Possible? The reported latency on the master from Sonarworks was 10ms. I'm pretty sure that the total system latency was 1054 samples which is only 10ms.

Maybe the player was just really ahead of the groove? Anyway, I'm pretty sure at this point we're just gonna nudge it back until it feels right and then I'm gonna let the dude fix his track. He was going to finesse the performance on his own anyway so here we go!