Reminder : the IMD, or Inter Modulation Distortion occurs when two frequencies f1 and f2 are played at once, and consists in the presence of two additional frequencies f1 + f2 and f1 - f2.
For example, while playing a 10000 Hz frequency together with a 12500 Hz frequency, IMD may produce two extra tones, the lower one of 12500 - 10000 = 2500 Hz, and the higher one of 10000 + 12500 = 22500 Hz.
It sounds like a dissonant interference sound playing together with the melody with the same rhythm, but with completely different notes.
Now that we have understood this, we can look at the spectral analysis of the three above audio examples.
First video, the part at 0:15 s (timestamps are not the same as in the video)
We can see nothing else than steady tones, noise, and harmonics H1 and H2 of the high singer. IMD would have produced variable tones different from H1 and H2. The detection threshold is -150 dB / frequency band.
There is no IMD in this recording. Any heard IMD comes either from the playback device, or from our hearing.
Here is the sonogram of the second video at 2:25 (timestamps are not the same as in the video) :
Here, IMD between the two singers is visible in the recording. I have indicated the f2 - f1 IMD that I can hear, near the bottom.
We can also see the f1 + f2 product as a copy of high singer H1, just below high singer H2, but with the same slope as H1.
Isolating a small part in the middle of the above sample, we can estimate the order of magnitude of the distortion.
The IMD that is present in this recording is somewhere around -53-(-23) and -53-(-30) dB, which is -30 and -23 dB, or, in percent, 3% and 7%.
Mind however that in a spectral view, dB are not the amplitude of the signal, but the amplitude in each frequency band. Let's say that we are somewhere between 1 % and 10%.
The IMD that I can hear in this recording is about the same as the IMD that I can hear in my own ears when I listen to the other recording that has no IMD itself. Thus it agrees with the 7% figure given in the beginning of the discussion for IMD in the human ear.
Here is the spectrogram of the last video at 1:03, with the three recorders playing at once (timestamps are not the same as in the video)
I chose a part where I can clearly hear IMD while one recorder is playing an ascending melody of three notes (the three ascending red steps), as I hear a descending pitched distortion (whose frequencies equal the decreasing gaps between the red lines).
Here, we can see that this IMD is not in the recording.
There are two suspect frequencies visible below 250 Hz (green lines), but they can't be IMD, because they would change as f2-f1 (the space between the red lines) changes.