Question for the Sony guys: Does a very small amount of "Time Stretch" cause ANY audio degradation?

riredale wrote on 2/25/2004, 1:44 PM
Just did an interesting experiment. I set up two miniDV camcorders (TRV8 and VX2k) and two Sharp MD-MT15 Minidisc recorders side-by side. I recorded audio simultaneously on all four for 30 minutes. I did claps for sync at the beginning of the recordings, and sync claps at the end.

When later brought on the Vegas timeline, I was surprised to see that (1) both camcorders were frame-accurate over 30 minutes; (2) both Minidisc recorders were frame-accurate with each other over 30 minutes; but (3) the Minidisc recorders were 15 frames slower over that 30 minute interval.

This means that my Minidisc recorders will lose about a frame of sync for each two minutes of recording. Why???

It would be a simple enough matter to shrink the Minidisc audio track down to match the camcorder audio tracks. This would be a shrinkage of 0.03%. How does Vegas do this shrinkage, and would there be any artifacts generated by this very small amount of shrinkage?

I will next run a test using 110v power for a Minidisc recorder, in order to determine if playback speed could at all be related to supply voltage.

Comments

riredale wrote on 2/25/2004, 2:34 PM
Did a second analog capture to Vegas from one of the Sharp Minidisc recorders, only this time I removed the two AA cells from the Minidisc recorder and used 110v power.

The result: exactly the same output as before, and I mean exact. Over 30 minutes, there was NO observable timing difference between battery and 110v operation, not even a small fraction of a frame.

So the 15-frame difference between Minidisc and miniDV recordings is still a mystery.
RichMacDonald wrote on 2/25/2004, 2:41 PM
Man, that really sucks. I feel for you. Sorry, I cannot help with either the explanation or to answer how bad the audio deterioration will be.
Catwell wrote on 2/25/2004, 2:43 PM
Every digital audio device has to have a clock that sets the sampling frequency. Unless you have a master (word) clock that drives the frequency for all devices you will usually have a different sampling frequency. One might be at 48,000 and another at 48010. This is not a big deal.

I use a seperate Hard Disk recorder for my audio and it is usually 3 frames longer over 10 minutes than my GL2. I have found that I simply hold down the control key and drag the video out to match the audio and all is well.
jeremyk wrote on 2/25/2004, 3:10 PM
Interesting question.

I have a tiny Sharp MT-877 minidisc recorder, and it's slow by one frame every 8 minutes. What I have to do is remove one audio frame every 8 minutes by cutting the audio track and dragging the right-hand segment one frame to the left. My guess is that maybe the extra current used during record loads the clock and changes its frequency slightly. I dunno. I also have a non-portable Sony MDS JE-530, and it stays exactly in sync for an hour or more.

I tried shrinking a one-hour audio track by 8 frames (!) and noticed some horrendus low-level warbling in the rendered audio. What I DIDN'T try is changing the audio track without keeping the pitch the same, i.e., letting the pitch change imperceptibly. I suspect that would work without artifacts.

I agree, it's a big pain to have to deal with the sync problem. Good luck.
Jsnkc wrote on 2/25/2004, 3:13 PM
If you get really bored and want to try something, take the minidisc player, run the line in to your camcorder and record the audio. Then capture the audio from your camcorder and see if it syncs then. Or try a passthrough from the minidisc to your camcorder.
jaegersing wrote on 2/25/2004, 4:22 PM
I would expect the 0.03% to be negligible, as long as you allow the pitch to change with the time stretch. Resampling to maintain the pitch will introduce more artifacts, but whether these are audible will depend on the source material.

Richard Hunter
PeterWright wrote on 2/25/2004, 5:17 PM
My Mini Disc, a 2 year old Sony, samples at CD quality, 44,100 or whatever, whereas mini DV is 48,000.

Could that be related to the difference in length?

ibliss wrote on 2/25/2004, 7:04 PM
But it's still based per second, so in each second the MD will sample the audio 44100 times and the DV tape 48000 times, in other words it shouldn't make a difference - regardless of sample rate, 60 seconds should equal 60 seconds.

The errors are almost certainly down to clocking, though it's interesting that the two camcorders stay in sync being different models.
farss wrote on 2/25/2004, 8:52 PM
Here's a theory:

It's all the fault of Vegas being too accurate and recording in NTSC.
The cameras record at 30 fps but Vegas does the right thing and runs it at 29.976. The answer seems to be in the right ball park.

If the correct figure is 29.976 then that'd account for 7 frames in 30 mins but someone should check the EXACT NTSC frame rate in Vegas and my maths.
Jay Gladwell wrote on 2/26/2004, 5:17 AM
Thought -- May or may not apply here.

Frustrating as it is technically, the "issue" here seems to be moot. With rare exceptions, and I do mean rare, a single shot (from point in to point out) in any given program may be seven seconds or less. With the exception of "Russian Ark" or a surveillance tape, I can't recall ever seeing a single shot lasting for 30 minutes.

The point is, over the course of 99.999% of most programs, the 15 frame/30-minutes of video issue would not pose any sync problems.

I would be curious to know at what point (which minute, second, and frame) did the audio and picture go out of sync?

However, you have caused me to do the same test with our Marantz PMD670. I'm anxious to see the results!

J--
Chienworks wrote on 2/26/2004, 5:28 AM
Robert, i don't think that's it. NTSC's frame rate is 29.9700299700299700 ..., not 29.976. And, any camera that is above the Fisher/Price or Barbie™ level ought to be recording at this same speed too. What many other NLE's do is to round the rate to 29.97 which causes an error of about 1 frame every 9 hours, 16 minutes, 6.666 seconds or so. But, since this entire project is being done in Vegas then the frame rate difference shouldn't be an issue anyway.

As far as the audio recorder's clock being off, it *seems* like it should be off by the same amount when recording as when playing back. In other words, it should cancel out. If the clock in your recorder is sampling at 44,000 instead of 44,100 (just for example), then it should play back at 44,000 too and 1 minute of audio should still last 1 minute.

I wonder if it's more of a case of a consumer grade audio recorder occassionally 'dropping the ball', missing a sample now and then, and ignoring it. Of course, this would make the audio recording shorter than real time which is the opposite of what has been reported here. Hmmmm.
riredale wrote on 2/26/2004, 8:53 AM
Interesting thoughts. My assumption at this point is that something in the Minidisc recorders is making the internal clocks just a little bit faster when recording. I do know that the Minidisc process uses some sort of heating element on the side of the disc just opposite the magnetic head, which is why battery life is much shorter for recording than for playback.

In any event, if the process of shrinking the length of a timeline in Vegas is artifact-free (assuming a pitch change) then at least there is a way to fix this in post. And the Minidisc format is a really inexpensive yet very-high-quality means of recording audio, and the recorders are the size of a fat floppy disk.

Now, regarding the pitch change: if a semitone is a pitch change of 5.9%, then a 0.03% change is the same as, let's see, carry the one... about one-half of a cent. In other words, no one is going to hear the difference.
Former user wrote on 2/26/2004, 9:03 AM
This might be interesting to try. Record the audio FROM your camera to your MINI recorder, and see if it stays in sync.

Dave T2
plasmavideo wrote on 2/26/2004, 9:03 AM
I've seen this exact question raised in several othe NLE forums. I think it is related not to the clocks within the devices, but is probably due to the fact that the audio from the MD was captured at 44K rate through a sound card (I'm assuming that is how you captured the audio) and the sample rate from DV was 48K (or 32K). Putting the 2 different sample rates on the timeline together probably result in the error, due to the small differences in how the 44K and 48K sample rates are handled.

If your sound card or audio software has a 48K digitizing option, try capturing it that way and see if they now are in sync.

Interesting science project - I might try it tonight with my MD recorder.
Catwell wrote on 2/26/2004, 10:53 AM
When I transfer the audio from my HD recorder, which is recorded at 48K 24 bit, I put out AIFF fils onto a CD and simply bring the file directly into Vegas. I still get the 3 frames per 10 minute difference.

It is important to maintain sync over a longer time. I am recording music recitals and the typical movement is between ten and twenty-five minutes long. I use two or three cameras and want to keep the whole thing in sync from beginning to end so I can just select any video track and it will be in sync with my master audio from the HD recorder. i would rather not have to resync for each clip in the final product.
Former user wrote on 2/26/2004, 10:58 AM
The only way you are going to guarantee sync between several cameras is if you use a SYNC GENERATOR and have all the cameras locked to it. OTherwise you are at the mercy of the mechanics and electronics of each camera.

Dave T2
Catwell wrote on 2/26/2004, 11:11 AM
Dave T2,
You are absolutely right. The only way to insure sync is to lock everything together. However, I don't have the funds for that kind of equipment so I do the best I can with what is available to me. By putting all tracks on the time line and sync the heads and then stretching the tails to sync them, I can get away with using my non Pro equipment. I don't have perfect sync but I can get close enough.
Chienworks wrote on 2/26/2004, 11:18 AM
Catwell, i think in your situation i would leave the audio as is and stretch the video out to match it. If you then disable resampling what will happen is that the extra frames will be generated by duplicating occasional frames wherever necessary. Since this will only happen a few times at widely spaced intervals it should be entirely unnoticeable and not incur any additonal rendering time to speak of.

If it does slow down rendering greatly, or if you wish to choose your own spots to duplicate the frames instead, you can manually split the clip in as many places as needed (1 split per frame that needs to be made up) wherever the action is unimportant (between songs maybe), drag the split parts over to the right by a frame, then extend the previous part by a frame. Same net result at your choice of locations, and a straight DV render will still be a straight DV render with no resampling considerations at all.
Catwell wrote on 2/26/2004, 11:27 AM
Chienworks,

I do indeed stretch the video to fit the audio. For me the most important part is the audio.

I had not thought of adding the frames myself. It makes a lot of sense to avoid the extra rendering time. Thanks for the suggestion.
BrianStanding wrote on 2/26/2004, 1:21 PM
Besides I had the identical problem with MD / DV sync when I was using Premiere. In fact, with Premiere it was worse, because unlike Vegas, there was no way to move the audio in increments of less than a frame. Resyncing was a nightmare!
BrianStanding wrote on 2/26/2004, 1:41 PM
I've had this experience, too.




Sorry you've been plagued with this, but I'm glad to see that it's not just me or my unit. I also routinely capture MD audio through the analog outs into my sound card and digitize at 48,000 / 16bit, same as my PD-150 audio. I've even tried capturing the MD audio over firewire through a media converter as a DV 48/16 AVI file. No matter what, I still get some degree of drift over 30 minutes or so.

I now only use my MD recorder for short (under 15 min.) stretches if I care about synch. Otherwise I use a backup DV camera, or wireless mikes. For the record, I've NEVER had a problem with two DV cameras going out of sync, even for stretches of an hour and a half or more.
farss wrote on 2/26/2004, 2:05 PM
Sorry about my poor maths and incorrect frame rate.
However the theory still holds. If the video and its locked audio was recorded at 30 fps and Vegas plays it back at 29.9700 then it will run faster, with me so far. By my calcs it would be out 54 frames after 30 mins: 30mins = 1800 secs = 54,000 frames x .03 / 30.


But the audio from the Minidisk isn't locked to anything so it'll stay at the same speed and hence eventually lag behind the vision. Now I'm not suggesting I'm right BUT from the experiment it seems two quite different cameras ran very close to holding sync over 30 mins and the two minidisc also held sync withe each other but something different happened to the video than what happened to the audio recorded by itself.

I'd point out that AE runs at 30 fps and that has caused problems bringing footage from it into Vegas.
ibliss wrote on 2/26/2004, 2:11 PM
Surely if the video was recorded at 30fps and played back at 29.97 then it would run slower?
Jay Gladwell wrote on 2/26/2004, 2:18 PM
Yes, sir. Same results with the XL1s and the Marantz PMD670. After 30 minutes of recording--both digital--there was a 15 frame lag from the Marantz to the XL1s!

Richard, I'm curious. What are you video taping that requires 30-minute takes?

J--