max@uwm.UUCP (Max Hauser) (05/06/91)
In article <11521@uwm.edu>, Paul S. Winalski writes: | | [The DCC] format performs sophisticated data compression on the signal ... | The format doesn't record the entire digital signal, but allegedly the only | information lost is stuff you can't hear, anyway, so it doesn't matter. | | I suspect that the golden ears, who already bash RDAT and CDs ... | are going to have a field day with [the DCC]. Yes, and they may even be right. But certainly not on the basis of "technical" arguments, if these fit the usual standards for that crowd; and similarly, they'll probably abjure any serious (objective) listening tests as well. If so, then even if the golden ears are actually right, they won't know it -- they certainly won't deserve any credit for it. Now in article <11525@uwm.edu>, Ted Grusec adds: | | ... the results are very impressive. Most listeners cannot tell the | difference between an original CD and the PASC encoded version, even | though the latter reduces the bit rate by a factor of 6! Very reasonably so. Perceptually-effective data compression is one of the prime missions of communications engineering over the last 50 years. Unlike copy protection (whose motivation is very different), it's been a vigorous and serious research topic not only for audio but in many other contexts. It is not a "new" topic and we should be wondering about its *absence* to date in consumer audio, rather than its presence now. You are actually experiencing human-interface data compression right now. You are reading a transcription of a spoken language encoded in print at around 120 bits/second. The same information, spoken, at telephone fidelity, requires roughly 104,000 bits/sec encoded with linear memoryless quantization or 64,000 bits/sec with US standard (mu-255) nonlinear quantization 16,000 bits/sec with ADPCM or sub-band encoding 2,400 bits/sec with US standard military (LPC) encoding 800 bits/sec with aggressive homomorphic encoding 120 bits/sec text at 20 equiprobable chars/sec of six bits each 20 bits/sec with aggressive Huffman source coding (exploiting relative frequencies of character groups) Obviously, additional cues are present in the various audible forms (you can't recognize my voice right now -- though you may recognize my testy style :-). But you see the point: the data rate required for a given core of content depends radically on exactly what steps are taken to remove redundancy -- redundancy that is generally at its maximum possible value when an analog source is encoded with brute force, i.e., constant-rate linear memoryless quantization (as in, for example, a CD). Max W. Hauser {mips,philabs,pyramid}!prls!max "The difference between an icicle and a red-hot poker is really much slighter than the difference between truth and falsehood or sense and nonsense; yet it is much more immediately noticeable and much more universally noticed ..." -- A. E. Housman Copyright (c) 1991 by Max W. Hauser. All rights reserved.