Can you pass this audio test?

There’s an epic debate going on in GQ over the authenticity of a viral video. There’s a lot of sub-debates going on within the thread, one of which concerns the perceptibility of sound delay. brazil84 seems to think that a delay of 40 ms is noticeable, while others contend that anything less than 80 ms is impossible. This thread is designed to test his hypothesis.

Below are 5 Youtube videos. Two have a delay of 40 ms and 80 ms respectively, while two others have advanced 40 ms and 80 ms respectively. One is the secret master video with the audio synced up correctly. Can you find the master?

VIDEO 1

VIDEO 2

VIDEO 3

VIDEO 4

VIDEO 5

Let me weigh in with my own thoughts here:

I have extreme bias since I know which video is which (and I’ll fill you all in after enough folks have voted), but I do detect the differences even at 40 ms. Even with my bias, I think that this test is slightly unfair because we will all be scrutinizing the videos. If I was just casually watching, I might not have noticed any delay at all.

And for those of you who are curious, more than an 80 ms delay is quite perceptible.

Thanks to everyone who participates!

My impressions:

1 - audio ahead, 40 ms
2 - audio delayed, 40 ms
3 - audio delayed, 80 ms
4 - audio ahead, 80 ms
5 - synced

My trick to figure it out:

I narrowed it down to 2,3, and 5, which I believe are +40, 0, and -40 irrespectively. Then I did them in pairs and the biggest difference between them I figured would be +40 and -40 so the outlier is the synced one. Well I still couldn’t tell very well so I chose 2 as I figured you wouldn’t put the synced one as the middle choice.

Dang, I really can’t tell.

I tried scrutinizing by concentrating on particular delay/advance markers, and sometimes seemed to think I could detect the correct delay or advance–but then it would disappear. I can make all of them seem advanced or delayed depending on which marker I concentrate on–kind of like you can make a necker cube seem to be facing whichever direction you like by concentrating on the idea of it facing that direction.

I’m at a not-so-new computer right now, and the display and interface is introducing enough irregularity that I can’t make an accurate determination. Your test relies on a perfect display and a fast CPU, and won’t work reliably with any other.

ETA: I wonder if that is what Frylock is observing.

I hope that you still vote. Go with your gut. Just because you can’t define the difference doesn’t mean that you won’t intuitively know which one is correct.

I guess this might be the case. I’m running a Mac with OS X 10.5.8 and a 2.4 GHz Intel Core 2 Duo processor.

Try letting the video buffer for a minute and see if it’s still irregular.

I getting some weird, unbalanced view counts on my Youtube channel.

I want to remind you all to listen to ALL 5 videos before voting. If brazil84’s hypothesis is wrong then you won’t be able to tell the difference between the synced video, the one 40 ms delayed, and the one 40 ms advanced. I won’t spoil exactly what I’m seeing, but there is one video that shows a 0 view count -which doesn’t make sense given the responses in this thread, but yeah…

When I post the solution I’ll also post the view counts to the videos. I’d understand lower view counts for the 80 ms delayed/advanced videos since you’re all probably scrutinizing the other three, but I don’t want to have to throw out the poll results because of an error in my testing method.

But we’re testing a claim that the difference is easily discernible, not that it is discernible even if only at an unconscious level.

I agree, but I’d still be interested in the data. The qualitative comments in this thread are equally useful data that should be weighed into the final result.

I definitely watched all 5. I see weird counts too. The counts are 2, 25, 0, 1, 1.

Maybe the view counter doesn’t increment unless you load the whole video? I only needed 5 or 10 seconds worth to tell.

I have a lot of experience in online first person shooter games, and an 80ms (.8 second) delay is very noticable. I also played Rock Band a lot which does require precise timing. Rock Band feels “off” if you don’t calibrate your TV and audio.

I had a hard time choosing between 1 and 5, but 5 seems to be in sync.

#3 sounds closest to me. The others are pretty far off.

I picked 2. 1 and 4 have the audio advanced from the video, 3 & 5 have the audio delayed.

BTW, how did you create this video? I suspect the 40/80 ms offsets won’t be quite right all the time due to monitor refresh rates.

The original thread is that crazy baseball through the pizza box thread, right? What’s the specific audio/video sync event that’s being questioned? Because noticing the difference between audio and video can be pretty easy for some things, like a video of a speaking person’s face. Even a single frame difference between them becomes quite obvious.

Much like the thread views on the board, the view count on youtube videos updates irregularly.

Considering poorly synced audio/video is a common situation on YouTube, for whatever reason, and some beyond viewer’s control, I don’t think this test is going to be of much value. It’s a lot like making a highly compressed JPG picture then asking how uniform the sky looks.

I pulled the video from Youtube and put it into Quicktime. Then I used GarageBand to create the audio tracks (pulled from the master video). I used a 40 ms clip of empty sound and an 80 ms clip of empty sound to create the delays. For the two videos that have the audio advanced, I deleted 40 ms and 80 ms respectively and placed them at the end of the audio file. All files were done in the same document, so they all had the exact same start and stop time. I double checked my work by visually checking the distance between “clicks” in each track at the same random points (GarageBand can be configured to display all of this). I then deleted the audio from the master video using Quicktime, followed by a cut and paste to bring back in the newly doctored audio tracks.

It was a pretty huge tangent having to do with whether or not the audio was doctored. I’m honestly not entirely sure why it became relevant, but I found this particular issue of great interest.

When I have some time I’ll download my uploaded videos and double check if this is the case. The videos certainly appear identical to the ones on my hard drive.

  1. Audio 40 ms early
  2. 40 ms late
  3. 80 ms late
  4. 80 ms early
  5. in sync

[edit] But are we looking for the leading edge of the small dark grey section, or the main light grey section?

I could see the ones with big delays, but the ones with small delays looked all about the same so I really didn’t know. I just went with instinct.

I think it’s #4.