Thanks for the links.
The first showed that 3 of the 16 significantly picked the wrong answer. What does that mean?
The others were random.
The 2nd paper was, if I remember correctly, well discussed at AVS. None of the subject papers by themselves showed detection. Only when all the raw data was combined there was something.
So, I would think that too didn't really show detection and perhaps with the summation of the data they should have used a much lower p than 0.05, perhaps 0.01?
I'm not sure. I do know just by looking at the impulse response between 96khz and 44.1khz it's obvious 44.1khz causes severe temporal smearing. Human hearing is significantly more sensitive in the time domain vs the frequency domain, being able to detect difference as small as 8 microseconds. It's rare to hear a large difference between hi res and Redbook audio, but some recordings make it very obvious. Coldplay's clocks, for example have much tighter transients on the snare and cymbals, and it's one of the few songs I've been able to ABX. 24 bit is probably unnecessary for music. There's not much more than a 40dB range even with classical.
Film is another story. In a properly calibrated system, average volume of dialogue is 85dB. You've got 20dB of headroom, so the maximum is 105dB. You could argue that the average home has a noise floor of 40dBa, concluding anything below this level will be lost, but its not that simple. I've done a few experiments myself, and can easily hear 1khz at a level of 10dB, and 4khz at 5dB with a 40dBa noise floor. Considering the majority of sound power is concentrated in the midrange, this gives us a perceivable dynamic range of 100dB. 16 bit audio has a theoretical dithered dynamic range of 96dB, but this is a best case scenario, and that only works for pure sine waves, noise shaping does not give a true 120dB dynamic range, it only ensures that sounds below the noise floor aren't lost, making them louder.
Keep In mind that 5dB sound is rms, the rise and fall off the wave will cross over below -100dB. A -95dB rms signal (10dB in a calibrated system) in 16bit audio will likely suffer quantization errors for this reason.
It's also worth noting that every time you double the sample rate, you gain an additional 3dB of dynamic range, which is why DSD gets away using 1bit. If you use a sampling rate of 176.4khz, for example, with a 16 bit bitdepth you'd actually have about 102dB of dynamic range, at 196khz about 104dB.
Sent from my 5065N using Tapatalk