speakers for 24/96 audio

Al, Huge errors applied to the highest harmonics only will result only in small sound change. There will be small difference in sound of cymbals and perhaps in ambiance.
I use 16/44 and like it, but try to be educated about it. That's all.

Bob - Thanks for the link. I suspect that THD is a dominating factor at higher power. Noise issue itself is non-existent in my opinion because if I cannot hear anything in a silent room at full power (dead silent) I don't worry. Many amps with similar 80dB THD+N performance are showing -120dB noise floor on the other graphs. Also, small amount of noise helps to increase resolution - technique known as dithering widely used in photography.
I would be more concerned with THD and it doesn't look good.

I don't know what is relationship between THD and resolution but I suspect that resolution will still bring better sound. Another reason for that is quantization noise that is smaller at higher resolutions. DAC1 does very good job here by using sigma-delta converter that pushes quantization noise to higher bandwidth (oversampling).

I think that our hearing ability ends up slightly above 16-bit perhaps 18-20bits but I'm more concerned with sampling rate because low sampling rate in addition to phase shifts in steep low pass filters increases quantization noise (or size of square steps to make it simpler).

That video is wrong. It is showing a stair step signal which is NOT what the output of a DAC would look like. The output will be smoothed by a filter in order to eliminate all that horrible spurious high frequency signal from the stair steps. The output filter will remove the stair step and restore the sine wave so that the signals are much more alike - even a little above Nyquist - absolutely No need to got 100Hz sampling to properly render a 10Hz sine wave.

warning not everything you see from Universities is accurate.

Also it is WRONG to compare signals in this way. We hear frequencies NOT the waveform as presented graphically! The closeness of the waveforms as presented graphically is NOT a proxy for how close alike they will sound!

"I think that our hearing ability ends up slightly above 16-bit perhaps 18-20bits but I'm more concerned with sampling rate because low sampling rate in addition to phase shifts in steep low pass filters increases quantization noise (or size of square steps to make it simpler)."

It's statements like that, Kijanki, that make me wonder if you know what you're talking about. The width of the data word has nothing to do with our hearing ability. How many bits per word determines how many loudness levels there are. It's the sampling rate in KHz that determines how high the frequency response goes. You know that, right?