16bit vs 24bit, Rubbish or Truth? Options

[Bourne]

-

This post has been edited by Bourne: Apr 2 2008, 19:26

[cabbagerat]

a) in practice, the average person (and sadly enough, many audio engineers) these days has far more hearing damage than they did 10 years ago, due to noise pollution, poorly mixed/mastered recordings, intentionally distorted musical content, excessive compression, improper EQ (can you say "Smiley face"), and high volume volume listening.

Interesting. Do you have any evidence for this? I would have thought that improved standards on noise in the workplace would have decreased hearing loss among the general population. It is well understood that loud sounds can damage hearing, but I am not aware of any evidence that excessive compression (for example) can have the same effect.

It would be great if you could present your evidence.

These are the things greater dynamic range bring to the table. The loss of detail due to poor representation afforded by shorter word lengths occurs in the smoothness of the dynamic changes (waveform amplitude changes) in lower level signals/components of the recording.

No. It is well understood that, for properly dithered quantization the quantization error is not correlated with the original signal (see Oppenheim and Schafer, "Discrete Time Signal Processing" or for a good modern treatment, or W. R. Bennet, "Spectra of Quantized Signals", Bell Systems Technical Journal, vol. 27, 1948 for the foundations of the theory). Stating this mathematically, we define the error samples e[n]:

e[n] = Q(x[n]) - x[n]

where Q() is the quantization process and x[n] is the samples of the signal under test. When dither is used (or for self dithering signals, where the analog SNR is below is the quantization SNR) the signal e[n] is not correlated with the signals x[n] and Q(x[n]). So what does this mean in practice? It means that the properly dithered quantization process is an additive noise process - it is equivalent to adding independent noise with a particular spectrum to the original signal.

Another thing is this belief that more quantization causes "roughness" in the output signal. As you are well aware, the reconstruction process does not produced a stepped output - it produces a bandlimited (and hence fairly smooth) analogue output. The output of a good sampling/reconstruction process is not "rough" or "stepped" in any way - if anything it will be less "rough" because of the strict bandlimiting imposed on the process.

This doesn't mean that recording, mixing, mastering an processing at high bit depths are without merit. There are plenty of good technical reasons to argue for this, though, so using audiophile terms like "roughness" and resorting to incorrect interpretations of the process are not necessary.

When you start with more significant bits, you have the ability to end up with more significant bits in the final product. Merely playing back a PCM stream that uses 24-bit words doesn't imply that the extra bits are significant by itself. Many listening tests involve a normalized 24 bit recording and a normalized 16 bit recording that has been dithered and noise shaped from the same 24 bit source. There is far less audible difference between those two sources, assuming a good dither/noise shaping algorithm is used. However, that test is not the test that validates the value of recording, mixing, and mastering in the 24 bit domain.

Of course. Mastering, mixing and recording in high bit depths (24 or more) is a good idea.