Topic: Melding sources...Question

[poppag76]

Ok, So I have a M/S matrix and taperbryan has his 483 source for the NYE RAQ show.  I want to combine the two sources.  This will be a first for me and I'm wondering if anyone has any tips for doing this with two separate sources, i.e., getting them lined up so it doesn't sound like shit.  I will be using SF8.  Thanks in advance.  Any other tips welcome.

[Patrick]

Just like doing a SBD/AUD matrix.  Use one source untracked and track the other one by song, and line them up in SF.  Tedious, but worth it in the end.

Check this link with pics: http://www.sloppyart.com/wavelab_matrix/

[Gordon]

use montage in wavelab.  look at the link posted above.

[cleantone]

The biggest problem is the fact that these sources were recorded (I assume) with different word clocks. So if you line up the first note of the first song the two sources will vary slightly and eventually become audibly obvious. You'll get different phase issues and eventually an audible delay. This could be as soon as 30 seconds, or as long as many (?) minutes. It is a pain in the ass for sure. Beside that even on the same word clock combining two complex waveforms can cause all sorts of weird phase issues. Give it a shot. I'm just chiming in to let you know this stuff. I'm sure others can give you some good advice on getting it down. I know if you search enough you'll find past threads on this issue. Don't let my words discourage you too much. Might as well try it out for yourself.

[NJFunk]

Here is a pretty good description of how to synch the timings as close as possible in SF.

http://www.gatheringofthevibes.com/bored/YaBB.pl?num=1162332750

1. Load both sources into Sound Forge.   
 
2. Change your rulers to show samples instead of H:M:S.   
 
3. Find a common point at the beginning of both sources.  A snare hit or some kind of click works best.  It doesn't have to be at the beginning of the recording, just near the front.  the important part is that you can find the same musical  point on each recording.
 
4. Zoom in to that point on each source and mark the peak of the waveform.  Try to make sure you get the exact same point on each.  You can be off by a sample or two, but not too much.   
 
5.  Repeat this procedure for a common point towards the end of each source.  Don't go by the meters telling you where you are, go by the wave.  Try to be exact, and make sure you put a marker on each source.
 
6.  Now that you have the same amount of "time" marked out on each source, you can measure each in samples.  Select the music between the marks on each recording and note the number of samples in your two selections.  they will always be different. 
 
7.  Here comes the math.  So say for instance you have a one-hour set.  Lets say for example you have the SBD at 158760000 samples, and the AUD used 158765000 samples to record the exact same amount of music.  You can deduce that the AUD recorder had a faster clock, because it recorded 5000 more samples in the same time as the SBD.  So, you need to slow that clock down.  You need to change the sample rate and then resample.  Take the larger number and divide by the smaller to get the ratio of one to the other.  So 158765000 divided by 158760000 equals 1.000031494 (rounded for space here...).  multiply your sample rate (I'll assume 44100Hz) by your ratio, and you get 44101.3888854 samples per second.  This is the actual sample rate at which the AUD was recorded.
 
8.  Take the result of your calculation and round it up or down to the nearest whole, so in our example, we have 44101.  Yes, one sample per second can make a lot of difference when you're talking about an hour.  Select the longer of the two recordings (in our case the AUD) and use the process > resample option in Sound Forge.  Make sure you only apply the new sample rate and DO NOT RESAMPLE at this point!  So now your recording that was 5000 samples too long is being played back at a slightly faster rate that makes up most of the difference.
 
9. Next take the AUD you just modified and use the resample option again.  This time, you will actually resample to 44100 and not just change the rate.
 
10.  Recheck your marks.  The difference in sample sizes between the two should be much less.  It's pretty damn near impossible to match them EXACTLY, but if you do it right you can get it to sound pretty good.  In our example, the process reduces the defference from 5000 samples(about 113 ms) to about 1400 (about 35 ms).  This is because you had to round up/down in step 8.  If you're a perfectionist, you can go ahead and delete a sample every 113401 samples, but that'll take a week.  I know.  I've done it.   You might be better off cutting a couple samples out of the longer WAV at the beginning of each song.  You can do the math.
 
11. Once you get the two time-aligned,  Trim the beginning of each recording, measuring the same distance back from the first markers, and mix the two with either the Sound Forge mix function or put the .WAV files into a sonar project and play with it from there.  Some people like to have a 60-40 mix but if you have a multi-track editor you can experiment with different settings.

[poppag76]

Thanks for all the replys.  I guess my Saturday plans are all full now.   

[easy jim]

Check out the discussion about drift and time-scaling in this thread http://taperssection.com/index.php/topic,77942.0/all.html

Chopping one source and repeatedly re-sycning it does not correct drift at all (which comes from 2 sources generated from 2 separate word clocks), and it is more cumbersome of a process than initially time-scaling one source to the other. 

NJFunk's posted description sounds like the right approach to me not knowing SoundForge at all.  However, I will emphasize that accuracy to the highest possible level is critical when time-scaling.  I would not round off at all if it is not absolutely necessary if you are calculating the actual variation in samples.  Even slight, milisec inaccuracy may create phasing issues and comb filter effects.  In AudioDesk/Digital Performer, you have to do a manual, click-and-drag style stretch or shrink to one of the .wav files using the time ruler as your guide instead of mathematical calculations.  I find that being off by any more than a milisec or two is enough for phasing and comb filter type effectes to come into play when I do this.  I can generally get things to +/- 1 milisec of accuracy...I would always at least try for that.



If you want it to sound crisp when you're done, be as accurate as is possible when time-scaling.

[NJFunk]

Check out the discussion about drift and time-scaling in this thread http://taperssection.com/index.php/topic,77942.0/all.html

Chopping one source and repeatedly re-sycning it does not correct drift at all (which comes from 2 sources generated from 2 separate word clocks), and it is more cumbersome of a process than initially time-scaling one source to the other. 

NJFunk's posted description sounds like the right approach to me not knowing SoundForge at all.  However, I will emphasize that accuracy to the highest possible level is critical when time-scaling.  I would not round off at all if it is not absolutely necessary if you are calculating the actual variation in samples.  Even slight, milisec inaccuracy may create phasing issues and comb filter effects.  In AudioDesk/Digital Performer, you have to do a manual, click-and-drag style stretch or shrink to one of the .wav files using the time ruler as your guide instead of mathematical calculations.  I find that being off by any more than a milisec or two is enough for phasing and comb filter type effectes to come into play when I do this.  I can generally get things to +/- 1 milisec of accuracy...I would always at least try for that.



If you want it to sound crisp when you're done, be as accurate as is possible when time-scaling.

I don't think Soundforge allows you to enter fractional sampling rates, hence the need to round.  Additional accuracy could be gains by first resampling both sources to 96k or 192k or whatever the highest possible sampling rate is in Soundforge and then doing the process since the difference between 44100 and 44101 is a lot greater than the difference between 192000 and 192001.