Need Help: Timing test

beatmad wrote:

innerclock wrote:

beatmad wrote:
What this means is that, although the ear may not be able to detect timing variations in a sequence (the average person cannot hear differences of 4ms), they are perceived as changes in pitch. The ear is sensitive enough to recognize that something is changing in the sequence.

Hi Beatmad - the frequency change/phase offset issue when multiple sounds are overlapping while changing the relative timing position between two or more is indeed a reality and of course this is something everyone can hear. However, I must disagree strongly with your view that most people cannot hear 4ms rhythmic jitter against a tight reference. In very strict tests with no harmonic overlap (evenly spaced rim shots against 4/4 kick for example) I have witnessed first hand many people adjusting rhythmic placement of sounds using delay lines with values less than 1ms.

If people are working by ear alone in this sub-millisecond region in order to get the right feel using equipment up to the task of staying tight in the first place - 4ms is more than wide enough a gap to drive a bus through.

Sorry if this comes across as hostile - I find the notion that humans have poor rhythmic sensitivity coupled with the idea that timing perception under 5ms is psycho-acoustic just puts everyone back in the dark ages and gets us absolutely nowhere.

Regards - David
www.innerclocksystems.com

Without going too far into it, the study I'm referring to centered on the ability to consciously detect deviations in a sequence, not with referenced sounds. Here's a quote from the abstract:

I understand your point - HOWEVER - picking small deviations in a regular pattern of events made by a single sound (no other precision rhythmic reference) at a fixed tempo does not really have any value as a test. Only in the presence of a reliable tempo/event reference can sub millisecond variations be heard clearly. In a musical/rhythmic/sequencer sense - this is a far more important and valuable way to measure things.

Regards - David

Are all MPC's rock solid in their timing?

Waxdatass wrote:Are all MPC's rock solid in their timing?

I've only tested Roger's vintage signature duo. After you've used either of those getting any other MPC seems like a backward step in many ways despite additional 'features'. There is just something about them. Like a solid full stop at the end of a sentence. Love for someone to post some numbers though. Be a shame if Akai didn't leave the timing engine alone in later generation machines - then again, far worse has be done throughout human history in the name of 'progress' after all....

best as always - David
www.innerclocksystems.com

innerclock wrote:Hi all - don't want to get bogged down in the details but check my litmus page halfway down for the MPC-3000 numbers if you need proof:

http://web.webhost4life.com/innerclock/ ... ge&name=34

You can put beats anywhere you like in the 3K but they'll always playback bolted in the same spot YOU put them.

Not so with many other beatboxes and PC/MAC midi sequencers.

You put the beats in Ok, you can even quantize the hell out of them.

Do the Litmus test on what you hear for yourself - hard 16th quantise comes out like a loose rubber band - why?

Hardware and Software not keeping things bolted down.

Badly written OS running inside compromised hardware = sloppy event triggering.

They got it right in the 60 and the 3K thats for sure.

Best - David
www.innerclocksystems.com

I once emailed Roger Linn about this. Here's his response:

The tight timing of the MPC is due to the fact that there is zero delay
between the sequencer and sound generator. It's all one big software
program, so when the sequencer part of the software says it's time to
play a note, the sound generator part of the software does it
immediately. There's no MIDI delay, no delays due to computer OS timing
priorities, no delays due to MIDI interface timing irregularities, and
no delays due to an external sound generator's software responding
slowly to MIDI.

There were no special tricks in the MPC software for specific drums. The
code was simply written so that sounds are played exactly when the
sequencer says that are to be played.

I hope this answers your question adequately.

Best regards,

Roger Linn

He also said that though many computer-based sequencers advertise high clock resolutions, they are not accurate or consistent. Even though the MPC is "only" 192ppqn, it's a rock-solid 192 rather than a badly-done 480/960 etc.

Just tested my MPC 500 with some interesting results. Using Innerclock's methodology and testing over 8 bars, the 500 is remarkably consistent with generally only about 1 - 2 samples drift. I say "generally" because there is the occasional 33 sample dip usually on the last quarter-note in about 2 bars out of 8. Where this occurs is not consistent - I'll have to test some more.

The other issue is that at 120bpm, the 500's quarters are 22081 samples apart - in other words not 120bpm - more 119.84 to be exact. I had noticed this before because on my system the 500 is useless in MIDI clock mode - the feel disappears completely. So I've been dumping beats free and editing them after. Testing using the above method revealed that the performance in MIDI clock mode to be pretty ropey - up to 64 samples drift. This may be a quirk on my system (MIDI via USB/OS X/EDIROL PCR).

If Akai could solve or fix that bpm mystery I would be very happy...

Well since I use FL studio 7 quite a bit I was interested to know how it would fair in David's litmus test, and I have to say the results were quite staggering.
Now I performed this test in four different ways, but all were for 16 recorded "clicks":

1) using the step sequencer then recording the audio.

2) using the step sequencer then rendering the audio.

3) playing into the piano roll, then hard quantizing notes a 100% then recording the audio.

4) playing into the piano roll, then hard quantizing notes a 100% then rendering the audio.

Now methods 1, 2 & 4 gave me absolutely flawless results of 22050 samples every time, (not really to surprised about the "rendered" audio.)
Yet method 3 was perfect up until it got to the "clicks" between beat 8 and 9 and there was an error of 256 samples, (thats pretty BIG!). But after that, back to 22050 every time.

I'm sure for most people who make beats in FL, this is great news as they will most likely render out there tracks anyway, but this may bring about an issue of hearing funny stuff while your creating.

As a side note I think that software could be adversly effected by a PC not properly tuned for audio/midi use. For me FL was audibly off timing before I knew how to tweak my PC.

I had a 60 and it was looser than the XL but thats a good thing, in fact id say that it makes the music sound less like it was programmed by short ciruit

Anyone test a 1k with JJ OS 2 on this thread yet?

Interesting material in this thread

Nice!

kneebone77 wrote:Anyone test a 1k with JJ OS 2 on this thread yet?

i just did the test with a 1000 with JJ OS 1.

hmmm.....


at 120 BPM things were almost always consistently solid at 22048 samples.
i never once had it at 22050 samples between clicks. always 22048...

any changes in length from 22048 were either a jump up to 22080, or an occasional drop to 22016.
no other values, just those....

any consecutive inconsistencies in clicks were always 22080 first, then 22016, then either back to 22048, or 22080 before resuming at 22048.

running this test a couple times i noticed twice that at bar 13 it would jump up to 22080, down to 22016, then up to 22080.


so while not often... maximum variation between any two consecutive quarter note intervals: 64 samples




------------------------------------------------------
i guess i should mentioned i hardly ever noticed any problems within the first 8 or 16 clicks, so i did this test over 20 bars which is when i started to see some minor changes in sample length.....
----------------------------
also: while eyeballing the waveform, i was tabbing to transient in REAPER to split and measure....while zoomed completely in this seemed accurate to me, but this might have possibly added another variable to the test...just thought i'd throw that out there....
-------------------------------------------

Thanks, punchdrunk...
Seems good enough to me!