That particular SOS article deals with latency. The definition of latency is the time elapsed between user input (on a key or drum pad) and device output (either MIDI note, audible sound, or both). Latency is an absolute measurement of time. However, the numbers I'm reporting below are relative time measurements - slightly different. I'm measuring how much a simple quarter note MD pattern shifts around compared to when it should precisely occur if the MD had a perfectly metronomic sequenced playback - in other words, sequence timing "deviation" or "slop". The reason I'm showing the three different scenarios is to distinguish between two different sources of deviation: 1.) MD internal MIDI sequencing deviation 2.) MD internal synthesis engine deviation The issue that Dave (innerclock) brought up is that although it is difficult for us to hear quarter note ticks shifting a few milliseconds back and forth in time on the MD, it is possible for him and many other people to hear the difference between the MD and other sequencers (i.e. older MPC and Roland sequencers) which had more rock-solid metronomic timing than the MD. So in conclusion, what's being evaluated here is not the MD internal latency (how fast it generates its sound output) - it is the consistency with which it generates that sound output. I am comparing the MD to other comparable MIDI sequencers which are more consistent in their sequenced sound output (i.e. have less deviation / slop). As an aside to the issue of deviation is another phenomena which I'm not sure anyone else has touched on. I hinted at this in my measurements of the EMX but wasn't quite sure what to make of it then. I realized after playing around more and listening critically, that there is maybe something to it... the EMX has significant deviation / slop, indeed a bit more than the MD. But, the EMX deviation numbers show a very consistent pattern of long-short-long-short etc. If I could put this into my own words, a drum pattern on the EMX somehow sounds "faster" to me at the same tempo than the MD when they're perfectly MIDI synced. Again, I think there is something to that very regular repeating periodic oscillation of deviation / slop that must cause this perception, because I like the way it sounds. I wish I had an old MPC to sample these boxes into to compare these sensations of a sampling sequencer with minimal devation / slop, one nearly perfectly periodic like the EMX, and one a bit more aperiodic like the MD, and one heavily aperiodic like the RS7k. It would be interesting to hear the same sounds and patterns played back through all machines. I digress... time to sleep. ;-) ----- Original Message ----- From: Jesse To: elektron-users@yahoogroups.com Sent: Wednesday, April 18, 2007 4:30 PM Subject: Re: [elektron] timing Great article Ripe, thanks for posting it... Final Thoughts Whether you've followed all the theory here, or just cherry-picked the results that interest you, there should be plenty of food for thought. Although many musicians rely totally on the latency figures reported by software utilities, the truth can be considerably more complex, involves many more factors than the size of the soundcard buffer, is sometimes prone to being slightly misreported, and is sometimes totally at odds with reported values. ----- Original Message ----- From: John Kimble To: elektron-users@yahoogroups.com Sent: Wednesday, April 18, 2007 3:28 PM Subject: Re: [elektron] timing How are you measuring the internal MD timing? just curious... Also maybe of interest for the issue in general: http://www.soundonsound.com/sos/Sep02/articles/pcmusician0902.asp cheers ripe ----- Original Message ----- From: Scott To: elektron-users@yahoogroups.com Sent: Tuesday, April 17, 2007 8:32 PM Subject: Re: [elektron] timing Correction to my last post below: my numbers are WRONG. I used an incorrect conversion method. These should be the correct time data points from my MD tests: MD triggering itself (MAXIMUM): 97 samples ~= 2.2 ms MD triggering itself (AVERAGE): 64.428 samples ~= 1.461 ms MD triggering MC-80 (MAXIMUM): 90 samples ~= 2.041ms MD triggering MC-80 (AVERAGE): 30.533 samples ~= 0.692 ms MD triggering Nord (MAXIMUM): 64 samples ~= 1.451 ms MD triggering Nord (AVERAGE): 20.333 samples ~= 0.461 ms Here is the formula I used: T_deviation (sec) = Sample_deviation (samples) / Sampling_frequency (Hz) So for the max deviation, for example: T_deviation (sec) = unknown Sample_deviation (samples) = 97 samples (as measured in Sound Forge 6) Sampling_frequency = 44100 (file recorded at 44.1kHz) Therefore, T_deviation = 97/44100 = 0.0022 seconds = 2.2 milliseconds I also double-checked this conversion in Sound Forge (even though Sound Forge truncates after the thousandths decical place in the time display) and it checks out OK. Therefore, I should change my statment of the MD triggering external MIDI devices: I think the average deviation between quarter notes is minimal if the external MIDI device has a tight internal engine, because a half-milli-second of "slop" is below the human hearing threshold. However - the maximum deviation of 2.2 ms of the internal MD engine is definitely within the realm (although definitely at the hairy edge) of human perception limits. However, all of this does not cover a more realistic scenario where the MD is playing multiple MIDI notes over one MIDI cable simultaneously, or playing multiple sounds internally. Whether or not I feel like going and doing those sorts of measurements... not really. That's not really my job. ;-) Though I hope someone out there is taking this as seriously as I am! Sorry for any confusion caused by my initial message. Scott ----- Original Message ----- From: Scott To: elektron-users@yahoogroups.com Sent: Monday, April 16, 2007 11:43 PM Subject: Re: [elektron] timing Exactly. You would think that would be the case, but things are not what they always seem sometimes. The disenchanting part is that the timing gets better when using the MD to trigger other devices over MIDI. This suggests, as I mentioned, that the slop is not only due to some random bug in Elektron's sequencer code, but it's also due to the internal synthesis engine of the MD. This leads me to speculate this could be an impossible situation to improve to a level where it would match devices like my MC-80 or Dave's MPC. Here are some more interesting numbers. Instead of looking at "maximum deviation" of samples between quarter notes, consider the average deviation: MD triggering itself: 56.375 samples ~= 19 ms MD triggering MC-80: 30.533 samples ~= 10.5 ms MD triggering Nord: 20.333 samples ~= 7 ms Ironically enough, I recall setting my "MIDI offset" in Ableton Live to right around 19ms when I had Live synced to the MD. ----- Original Message ----- From: Patrik Rydberg To: elektron-users@yahoogroups.com Sent: Monday, April 16, 2007 3:37 AM Subject: Re: [elektron] timing Scott skrev: > > > MD original triggering its own internal rimshot: 97 samples / 33ms (once > every 4 measures) > MD triggering the MC-80 rimshot: 90 samples / 31ms (once every 4 measures) > MD triggering the Nord pulse: 64 samples / 22ms (once every 4 measures) > Interresting... > Returning to the MD, however, reexamining the numbers: > > 33ms internal MD sound > 31ms external MC-80 sound > 22ms external Nord sound > > It seems there could be as much as 33% of the MD's overall timing slop > could be due to the synth/sample playback engine, whereas the remaining > 67% could be due to the sequencing engine. This is by no means an > exhaustive scientific study, however, a preliminary one whose results > are a bit intriguing... Well, the way I see it, if the Machinedrum is running on internal clock, triggering its own sample engine from its own sequencer, the timing has every possibility to be very accurate, not just as good as triggering external synths via midi. Using MIDI, you have all the problems with the midi protocol. Internally, between the MD sequencer and synth engine, why use midi? Compare the situation to a DAW. When triggering external midi stuff they are sloppy, but when triggering VSTi:s the timing is sample accurate. /Patrik www.cutpaste.org [Non-text portions of this message have been removed] [Non-text portions of this message have been removed] [Non-text portions of this message have been removed] [Non-text portions of this message have been removed]
Message
Re: [elektron] timing
2007-04-19 by Scott