The way digital audio quantization is covered in textbooks is flawed. A waveform derived from an undithered digital recording at 16 bits is treated as if were the same as an analogue signal with signal to noise ratio of 96 dB. This is actually wrong and leads to false conclusions. Adding 1 millivolt of noise when making an analogue recording would not make all details below 1 millivolt in height inaudable, using a quantization level of 1 millivolt when making an undithered digital recording would.
Put simply the signal to noise ratio of 16 bit audio is not 96dB in the conventional meaning of noise unless it is dithered. As for false conlusions well for example using more than 16 bits should not be dismissed by using a signal to noise ratio argument based simply on the number of bits.
... Adding 1 millivolt of noise when making an analogue recording would not make all details below 1 millivolt in height inaudable, using a quantization level of 1 millivolt when making an undithered digital recording would.
Which is why you should always dither. Most literature assumes dithering, I guess. And to be precise: by quantization level you mean "quantization threshold", right? i.e. quantization level = (quantization step-size) / 2?
Chris
The way digital audio quantization is covered in textbooks is flawed. A waveform derived from an undithered digital recording at 16 bits is treated as if were the same as an analogue signal with signal to noise ratio of 96 dB. This is actually wrong and leads to false conclusions.
Every textbook derivation I've seen has made clear that this is only true under the assumption that the quantization error is uncorrelated with the signal (e.g. that the signal is dithered). Which text are you using?