Topic: Matrix Time Drift -- How Much Is Too Much

[junkyardt]

Im skeptical about Vegas routine...without knowing what the stretch is really doing...I hate not knowing.

I like the math approach...

  as far as i knew, you're stretching it either way. i mean, the math is just a way to determine how much to stretch by, correct? and what do you mean 'you don't know what it's really doing'? it's fairly simple, all it's doing is speeding up or slowing down the recording. you can have it change pitch if you want it to, but you're in control of that in Vegas. if you don't want it to change pitch, it won't.

[runonce]

Im skeptical about Vegas routine...without knowing what the stretch is really doing...I hate not knowing.

I like the math approach...

  as far as i knew, you're stretching it either way. i mean, the math is just a way to determine how much to stretch by, correct? and what do you mean 'you don't know what it's really doing'? it's fairly simple, all it's doing is speeding up or slowing down the recording. you can have it change pitch if you want it to, but you're in control of that in Vegas. if you don't want it to change pitch, it won't.

"Stretching" is just the name of the function...what does it =technically= do? Is it a resample routine?

Someone else asked - how do we find the correction factor when the recordings arent the same length? (presuming two recorders started and stopped at different times.)

Is there a way to determine how many samples are contained in X seconds of music?

[ghellquist]

I think we are addressing a situation where there are 3 clocks involved.
1 - recorder A
2 - recorder B
3 - playback device
If there is any difference between 1 and 2 - it will result in a different pitch when played on 3 - correct?
I wonder we were able to use two discrete clocks during playback - would the sources be the same pitch?
Is there anyway to determine a given recording's "absolute" sample rate? (rather than the generic "44.1")

First, you might be able to determine the exact sample rate if you have a very accurate signal of known pitch recorded. Remember though that the crystal oscillator in you recording box is very accurate already, so you have to know the reference pitch with even better accuracy.

Secondly, you could vary the speed of the two playbacks. That is basically what you do with the "stretch" algorithm in software. It could be done in hardware, playing out the two signals to an analog signal and then perhaps converting it back to digital again. Much simpler to use the software.

Gunnar

[ghellquist]

Is there a way to determine how many samples are contained in X seconds of music?

Simple, even I know that one. X times the sample rate.

// gunnar

[ghellquist]

How does the addition work with the digital scale?   (I searched but couldn't find anything using my chosen keywords).  I normalized each of my sources to -6 dB.  If I mix them without touching either of them, what will the peak amplitude of the resulting file be (assuming two -6 dB peaks overlap)?  And if I increase one of them to -3 dB and reduce one of them to -8 dB?

Before anything else, you have to decide on which kind of decibels we are talking. Generally we would talk "sound pressure level", but some of the decibel scales are "power related". There is an excellent article with not too much math:
http://www.engineeringtoolbox.com/adding-decibel-d_63.html

First, it depends on how the two signals are correlated to each other. If they are very similar, or exactly the same, adding two signals together will yield a 6dB increase. If the two signal are exactly opposite as in phase inverted, the signals will ge totally nulled, no output at all. Generally we calculate with unrelated signals to increase somewhere from 3 dB to 4,5 dB. In a matrixing situation the signals will be close but not exactly same so I would count on -4,5 dB.

Secondly the dB scale is logarithmic. A -3dB added to a -8dB will (when totally correlated) increase the signal to about -2dB (or slightly above that, something like -1,75dB).

Gunnar

[it-goes-to-eleven]

Part of my trial and error is repeatability.  It's common to the device, therefor common to all media captured on the device.  Once I guess right, I should never need to guess again.
...
My Korg MR-1000 seems to differ between left and right channels.  Or at least the converted results via audiogate.  It's the same device so it always starts in sync at 0 to 0.  Basically 0.0006 seconds per hour difference between channels.

Hardware drift tends to vary with temp.  So for most people here, there is no constant/linear fix.

Your case seems to be different than most - if it is software drift in audiogate.  If it is constant, they could easily fix their conversion software.  Or it might be a hardware problem.  It is not an issue you should have to fix.  Have you tried contacting Korg about it?

[Shadow_7]

In my case I think it's mostly software related.  And by software I mean in the part of the flow that is AFTER audiogate.  Audiogate can get me CLOSE to my end goal.  But I'm also compensating for an HD camcorder that runs at about a factor of 1.00011 faster.  About 20ms per 5 to 10 minutes.  My korgs channels seem to be drifting 1.00000003 per hour.  Just enough to reach perfect phase inversion between 30 minutes and 60 minutes.  And maintain it for an equivalent amount of time.  Hearing it "correct" is apparently a new thing for me.

I'm not sure if that's audiogate or not.  But it is enough to affect the low end perception of my mics.  Since most times I'm recording 2 hours continuous per session.  It also makes gui editors undesirable because you have to wait for that content to load.  And otherwise be there for each step of the process.  All I'm using audiogate for is to convert from 5.6 DSD to 32/192.  (was 24/96).  I'll likely skip that step in future sessions by just recording in 32/192 (or 24/192, not sure which it's actually doing).  And that all so I can do the audio equivalent of oversampling  to the desired result.  Since I know ahead of time that I'll be adjusting speed. 

Sox was injecting drift AFTER using sox to correct the drift.  So it's kicked to the curb.  Currently making some strides with sndfile-*, but you seem to have to break everything up into individual steps.  No amping, concating, compressing, or other stuff all in one step.  The drift portion currently looks something like:

  sndfile-resample -by 0.229687507 -c 0 temp_L2.wav temp_L3.wav
  sndfile-resample -by 0.229687500 -c 0 temp_R2.wav temp_R3.wav

The rough equivalent of 44.1kHz from 192kHz.  -c 0 takes forever, so I ran a few test runs with -c 4 and results were satisfactory.  0.000003 percent -ish in audacity's version of the process for speed on the right channel.  As far as I can tell, audacity does the same conversion as sndfile-resample.  And falls back to sox if sndfile isn't present.  I'm pretty sure I need to file a bug report against sox.  Not sure if I should against audiogate.  sndfile seems to satisfy at this point.  Albeit the slowest of all conversion software to date (in part to the highest possible settings).  I was using sox to make sure all edits landed on solid seconds to make CD tracks land on/near 1/75th of a second.  My stand alone CD player wont play a CD if it doesn't.  At least not a 700MB CD.  It didn't seem to mind when 650MB CDs were used and SBE or whatever the technical term is were not accurate.

I'm not too worried about pitch.  These adjustments are so small it's not likely to affect intonation and stuff.  Bad intonation is generally measured in beats per second.  We're talking about shifting one low frequency wave form one crest per hour.  If you're adjusting pitch on that scale factor, you're probably injecting more problems than might be solved.  Or at a minimum increasing the processing time needed to perform the process.

[mattmiller]

I've got two sets from Friday night mixed and I'm happy with the results.  I can't hear any drift, which is good enough for me.  But I just noticed that the two sets are awkwardly DIFFERENT in file size.  The mixdown of set 1 is 65 minutes long and is 1.47 GB (48/24).  The mixdown of set 2 is 55 minutes long but is only 929 MB.  So set 1 is only 18% longer but is 58% bigger in file size.  Both still say "48000  32 bit  Stereo" in CEP.  Is this possible?  Is there some sort of diagnostic I can use in CEP to find difference between the files?

[Shadow_7]

When you speed up a track, it's going to get shorter.  If it matched another track before being sped up, then the other track will be longer than the one that got altered.

-----

When you have two tracks that start at different times, you need to find TWO sync points common to BOTH tracks.  The further apart the better.  Consider one of the sync points your new 0 (ZERO).  Basically a two step process, make them run at the same speed (no drift), then trim each to same sync point / point in time.  Bear in mind that you'll likely need to re-find the sync point (zero) after resampling.  Could be simple math and automated.  Could be going through the ID stage a second time.  Which you should probably do to verify results anyway.

-----

I agree it's not something that I should have to do.  But in my case the drift is so small, and the tracks / hardware so closely related, that any variance in temperature will likely not produce a variance in clocks and drift large enough to be of concern.  And probably not that adjustable anyway depending on the decimal point accuracy of the CPU/FPU (and software) performing the conversion.

I've got a process now and it seems to work.  sndfile-resample is my friend.  And audacity his good buddy.  I still need to find an equivalent to sox's trim function, manually punching in the timestamps is a bit tedious IMO.  Resampling happens first and relatively blindly / on faith.  That way the gui only loads small and manageable files.  I'm loving the results so far.  Finally something put the Tuba player back into the brass quintet.  I've tried all scores of EQ and other tricks to limited success until I figured this one.  I even tried a few different mics which generally proved more problematic.  I've listened to these things many hundreds of times and it's bugged me every time, until now.  It's like I got a $500 or would it be $5,000 upgrade.  Since it looks like I wont need to try those $3K+ mics in an effort to keep my sanity.  At least not anytime soon.

I've still got to track down what's the root cause.  It shouldn't be the hardware as that would be problematic when it creates new files after every 12:41.2??.  Lot's of A/B-ing in my near future I guess, to see if audiogate and only audiogate satisfies.  I've generally left it to audiogate to convert from DSD to an editable format, and that's it / period / end of story.  Everything else is done in other software.  So I don't know if it's the device and/or audiogate that is the root cause (yet).  I do know that sox's trim function when pulling content from about an hour in injects drift where there is otherwise no drift.  Could just be my current version of sox.  Could be my old 32 bit laptop.  Could be something in the file formats.  Could be many things.  Korg can wait until I figure out what it is and what it is NOT.

[admkrk]

I've got two sets from Friday night mixed and I'm happy with the results.  I can't hear any drift, which is good enough for me.  But I just noticed that the two sets are awkwardly DIFFERENT in file size.  The mixdown of set 1 is 65 minutes long and is 1.47 GB (48/24).  The mixdown of set 2 is 55 minutes long but is only 929 MB.  So set 1 is only 18% longer but is 58% bigger in file size.  Both still say "48000  32 bit  Stereo" in CEP.  Is this possible?  Is there some sort of diagnostic I can use in CEP to find difference between the files?

things like silence and solos will use less space than say 6 guys going to town at the same time.

[junkyardt]

Shadow_7, are you like, on acid or something? I've tried to make some sense of your posts, but they seem to be little more than you rambling to yourself.

[it-goes-to-eleven]

I've got two sets from Friday night mixed and I'm happy with the results.  I can't hear any drift, which is good enough for me.  But I just noticed that the two sets are awkwardly DIFFERENT in file size.  The mixdown of set 1 is 65 minutes long and is 1.47 GB (48/24).  The mixdown of set 2 is 55 minutes long but is only 929 MB.  So set 1 is only 18% longer but is 58% bigger in file size.  Both still say "48000  32 bit  Stereo" in CEP.  Is this possible?  Is there some sort of diagnostic I can use in CEP to find difference between the files?

things like silence and solos will use less space than say 6 guys going to town at the same time.

NOT.

[mattmiller]

I've got two sets from Friday night mixed and I'm happy with the results.  I can't hear any drift, which is good enough for me.  But I just noticed that the two sets are awkwardly DIFFERENT in file size.  The mixdown of set 1 is 65 minutes long and is 1.47 GB (48/24).  The mixdown of set 2 is 55 minutes long but is only 929 MB.  So set 1 is only 18% longer but is 58% bigger in file size.  Both still say "48000  32 bit  Stereo" in CEP.  Is this possible?  Is there some sort of diagnostic I can use in CEP to find difference between the files?

things like silence and solos will use less space than say 6 guys going to town at the same time.

NOT.

That's what I was thinking -- 48,000 samples per second, each described by 24 bits, should be the same no matter what they're capturing.

[Shadow_7]

Shadow_7, are you like, on acid or something? I've tried to make some sense of your posts, but they seem to be little more than you rambling to yourself.

Is bash an acid?

http://home.earthlink.net/~shadow_7/korg_resync.sh

That's basically what I'm doing.  My two tracks on my two channel field recorder drift.  Just to enough to operate out of sync 80% of the time.  Perceived by me as no low end.  Not bad enough for reverb effects.  But bad enough for doing the equivalent of inverting the phase of a stereo image and publishing that as a commercial CD.  I don't know if that drift is artificial (software) or part of the hardware.  But it's there, and it's plagued me for over two years, which is how long I've been recording with this device.  I suppose it does sound a bit like I'm talking to myself.  I should probably quote more often than I do, which would probably help.  But as far as I know I'm speaking english.  Not that I was ever good in it in school.  Anyway that script seems to correct the data issues for me.  I could take some screen captures if that helps you to understand.  Or maybe some sound files would work?

BEFORE:
http://home.earthlink.net/~shadow_7/rnsm.mp3

AFTER:
http://home.earthlink.net/~shadow_7/rnsm3.mp3

No brasses, but there are bass drums and timpani to indicate what is most affected.

[admkrk]

I've got two sets from Friday night mixed and I'm happy with the results.  I can't hear any drift, which is good enough for me.  But I just noticed that the two sets are awkwardly DIFFERENT in file size.  The mixdown of set 1 is 65 minutes long and is 1.47 GB (48/24).  The mixdown of set 2 is 55 minutes long but is only 929 MB.  So set 1 is only 18% longer but is 58% bigger in file size.  Both still say "48000  32 bit  Stereo" in CEP.  Is this possible?  Is there some sort of diagnostic I can use in CEP to find difference between the files?

things like silence and solos will use less space than say 6 guys going to town at the same time.

NOT.

That's what I was thinking -- 48,000 samples per second, each described by 24 bits, should be the same no matter what they're capturing.

maybe i'm thinking of flacs, but i don't see why this should be much different. less information is still less information.