Sample Rates and Running Time

jmpatrick wrote on 12/19/2001, 9:03 PM
I'm still working on my 3 camera live shoot.

The 8 tracks of audio were recorded straight to the H.D. at 24/96. The 3 camera angles were recorded on 3 Canon GL-1's. I mixed the 8 tracks of 24/96 audio to stereo, and render at 16/48 for DV. Then I bring this mixed audio into the project that has the 3 camera angles. I sync'd the clap at the start of the show no problem. However, by the end of the 50 minute set, the sync is gone. It turns out the mixed 16/48 audio runs about .250 of a second longer than the audio from the cameras. The only thing I can imagine is that the 24/96 to 16/48 conversion is stretching things very slightly. The next time I do a live shoot in this fashion, I'll try recording the 8 tracks of audio at 16/48. Maybe that will keep things locked.

Next problem: I've used the stretch feature to adjust the mixed audio so it stays in sync. It's easy to do, but the results are disappointing. There's an audible warbling sound. I've tried all of the settings...some are better than others. I just can't get rid of the warble completely. When I use SoundForge 5 to adjust the length, I also have problems. I can set the end length, but when I apply the stretch the end result is always off by a few milliseconds. What's the best way to stretch to an exact number?

jp

Comments

jmpatrick wrote on 12/20/2001, 4:13 PM
anyone?
pelvis wrote on 12/20/2001, 8:47 PM
Long answer...

Canons do no record at exactly 48k, and this is a drag. If the reasonable assumption is made, by the DV app, that the incoming audio is a "pure" 48, you'll wind up with a/v sync drift, (not offset, drift; a/v sync getting worse over time). I use a GL-1 so I've seen this exact problem many times in the early days of Vegas. Better Sony's do not have this problem and appear to record at true 48.

To correct potential a/v drift, SF DV apps resample ("re-align") DV audio with every DV video frame. This means a/v from the DV source will always line up with itself over time. This also means that (the bad part) you probably won't be able to line up Canon DV source material with other true 48k sources without some fancy footwork.

So, you could:

1) Use the audio captured along with the DV clip as a reference, and chop up the hi-grade audio to match

or

2) If the Vegas timestretch isn't good enough you can use Forge.

Set the Vegas time display to samples. This will be your "measuring stick". Measure the duration, in samples, between a visual clap mark at program start and some other visual clap mark at the end of the program. This duration is what you want to stretch the audio to.

So now find the clap mark in/out in the good audio track. Do a split of this event so the clap marks are EXACTLY on the head and tail of the event; trim edges if needed. Open this event as copy in Forge (right click the Vegas event and you'll get a menu option for this...)

Ok, now the audio is loaded in Forge. You can tell Forge to stretch this audio to the desired length in samples. There are a number of stretch options- pick the one that best suits you material. Save the thing out, it will be loaded back inb Vegas as a new take, and you should now have sync.

Painful, but it will work.


jmpatrick wrote on 12/21/2001, 6:19 AM
Good ideas Pelvis...

I'll give them a try.

Thanks!