The sound of the collapse of the south tower was recorded in the Naudet documentary, from their vantage point in the North Tower lobby. It provides another way to at least try and estimate the collapse time, as you can see (or rather hear) from the clip itself.
Click the image above to display the video, or right-click to save it to your PC (it’s about 2 MB). Play it for yourself, and see if you agree with the following.
- The rumble signifying the beginning of the collapse begins at the 8 second point, and gets louder from then on. The continuous rumble suggests to us that we're hearing the collapse, not preceding explosions, and if so then the tower probably began to fall at least a second earlier (it takes time for the collapse to become loud enough to be recorded, then pass through the air to the other lobby). However, to err on the side of safety we're recording 8 seconds as the collapse time start.
- At 14 seconds in we get what appears to be the first sounds of major debris hitting the north tower lobby. We can't say if that's true, or where the debris came from, but if it were in the region of the impact point then we'd expect it to have taken 8 seconds or more to travel through the air. This would put the collapse back to 6 seconds in, but that's too much speculation to be sure: we'll stick with 8 for now.
- The rumbling and sound of debris hitting continues to grow, despite the fact that the cameraman and firefighters are a couple of corners away from the lobby. The picture and audio both fade out at this point, meaning we can't tell when the collapse ends for sure, or even if it's recorded here. However, there is a point around 22 seconds when it seems like the rumble of the falling tower is replaced by the sounds of local destruction, flying glass and so on, so if we had to pick a plausible earliest collapse time end then it would be there.
Looked at all together, then, we have a collapse start time of 8 seconds, the end at 22 seconds, a 14 second collapse time in total (and with some reason to believe that's an underestimate). There's no way we can claim split-second accuracy, but then that's not the point. The aim here is to use this method as a check on the video estimates, and we think it clearly shows the 8.4 second end of the spectrum as very unlikely, while 15 seconds and over looks distinctly plausible. Download the file and try it for yourself.