The audio from this excellent video makes the sound of shots 1-3 sound very similar to shots 5-8. Yet, I had previously agreed with Chris Martenson that the sound of the two sets of shots were very different.
Is it simply that Crooks fired the second set of shots while holding his rifle sideways, using the side sights? Or could the difference in comparable sounds be the position from where the video was taken? Can we compare the audio of those shots from various positions on the field to speculate on where the second shooter might have fired from?
But similar spectral suppression does occur in a few other places. Granted, it certainly isnāt happening often, but here is one good example I picked out:
This is the third audio channel at 1:43. What is happening here is some fairly quiet talking coming out of the radio, and you can see that itās at a fairly low amplitude. However, briefly, all of the background noise above 4 kHz gets suppressed.
It appears that the speech here has a moment where there is concentrated amplitude at 700 Hz, and that seems to take energy away from the other FFT bins in the same time slice - the bins that were already weaker to begin with. With 16 bits of resolution, it seems like this type of thing should not happen, but it does. Why? My best guess is that it has something to do with the A/V compression being used for the recording. Compression causes artifacts on both video and audio. These artifacts are supposed to be so small that humans donāt notice them, but computer analysis can reveal them, right? I think these black areas are compression artifacts. @offtheback, it sounds like you have a lot more experience in audio analysis than I do, so what do you think? Could this be a result of the compression being done by the the cruiserās recorder?
Let me look when I get back to my computer. One answer may be that the source of the hissing noise is the radio channel, which, when it has actual signal, transmits that into the cabin instead of noise.
Audio and video coding rates shouldnāt be codependent.
And yes, frequency decomposition and selective frequency bin manipulation is part of some compression schemes, as the goal for compression is to remove inaudible sounds if you have to remove something. But this recorder doesnāt seem to be working that way, from lots of other examples.
Theoretically, you could clip a region of sound to a different track in Audacity, offset it, and then subtract it from the original track to illustrate this.
Here we need almost perfect phase alignment to the sample to make this work given the frequency content.
I think it may actually be doable. Certainly to within a few samples, and try several possibilities. I was timing the tracks manually to I think a plus/minus two sample error when I was investigating arrival time jitter as a fingerprint of manipulation.
The question we really want to answer is whether somebody chopped out some stuff and added gunshot foley. I confirmed that all 8 reports were so similar that they could be overlaid and the pressure waves (or whatever they are) were virtually identical, but I just donāt know if thatās expected or not.
I fully understand that, and thatās why I asked: you donāt have to do this in manually in audacity, but can do it in every programming environment able to read sound files. Iāll try to write a script.
I am referencing a theory, but I do not believe it is correct. However Chris mentioned it in a video due to the side sights visible on Crooks rifle. You can clearly see the side sights in this FBI photo of the weapon:
I took a look at the time stamp you referenced. Similar dropouts occur in the other channels for the same time. In channel 2, they are somewhat overwritten by clipping, which manifests as broad spectrum noise. In channel 1, you can see it overwritten with other noise sources that donāt drop out (horizontal purple traces).
So I think this is an artifact of the police radio, which is amplifying static except when it is notā¦
BTW, with regard to your TOA analysis, it assumes line of sight between source and recorders. But if someone is crouched on the roof of the building behind Crooks, I would say no one has line of sight to that sound source except for the podium.
I was scratching my head for the longest as to why the TMZ path length was slightly delayed for shots 1-3 as compared to what you would expect from the podium timing. But if the shots emanate from the area just behind Crooks on the next roof, there is a huge two-story building blocking the line of sight path to TMZ, and once the sound has to pass around it, the timing matches up.
If you are willing to program it, you can run a quick and dirty brute-force cross-correlation with a portion of the report 1 signature and the whole area of interest in the audio samples. We wonāt mess with the means and variances and just assume we can scale the results by the value achieved when the report 1 signature portion exactly aligns with itself. We would just need to define starting and ending samples of the clip we want to correlate with. Here is some pseudocode.
Shot 1 Starting sample == SS1;
Shot 1 Ending sample == ES1;
Length == ES1 ā SS1 + 1;
Sample range of original sequence == 0 to N;
Array of samples == Sequence[0 to N];
Output Array == Output[0 to M]; ** If array of samples is only N long, M canāt be longer than the original array minus the length of the shot 1 snippet plus 1. The audio here is really long, so just clip plenty past the last shot and select a reasonable M that will get you two seconds past the last shot. ā0ā could reasonably be about 1 second prior to the first shot **
Calculate scale factor
Scale_factor == 0;
For i == SS1 to ES1 {
Scale_factor += SQUARE(Sequence[i]);
}
Calculate cross-correlation
Step == 0;
For i == 0 to M {
Output[i] == 0.0;
For j == 0 to Length {
Output[i] += Sequence[SS1+j] * Sequence[i+j];
}
Output[i] /= Scale_factor;
}
And then plot the Output array. Floating point numbers should be between -1 and 1, but canāt swear that will be true as we have made a few approximations (like zero mean).
Thanks. I donāt think the video has been maliciously edited at this point. More likely is that the lower-resolution video is just that: lower resolution, with fine movements masked by the compression. For anyone interested, hereās a video explaining how mp4 compression works and how it recycles frames, which can mask small movements, especially at lower resolution:
Iāll say this and be blount about Gary at PT: you donāt need to be angry. He spread disinformation without a care in the world.
One, he tried to claim that a shot could not be fired from the 3rd window and clear the fence to reach Trump. He did that without any actual investigation or calculations to back up his claim. That is dishonest no matter how you look at it.
Two, he tried to claim that there were people in between the AGR windows and the fence at the time of the shooting. We all know that is a lie because multiple videos confirm it is a lie. So: two blatant dishonest statements from Ol Gary.
Donāt stress about Gary. He clearly has his own agenda and it isnāt the complete truth.
Could I trouble you for the actual original still frames that you used to make the enhanced, stabilized GIF? I would like to do some analysis and comparison but if there is a bunch of other processing larded on top it begs the question of what process caused what effect.
I already have still frames for the ālow-resolutionā version from Roger.
In the tmz video you can see a man filming himself moving from right to left in front of the camera.
Thatās probably what happened when the blonde woman bumped into the cameraman.
Thatās supposed to be this guy, but heās standing next to the cowboy hat guy the whole time.
