Stainless Steel Rat <[EMAIL PROTECTED]> writes:
> * Sean Buckingham <[EMAIL PROTECTED]> on Mon, 05 Jun 2000
> | Whilst I would have little technical comprehension of how discs could be
> | coded differently yet still read on the same machines, surely this is what
> | is happening?
>
> Not even close.
>
> What is different is the psychoaccoustic model used for the encoding
> process. The way ATRAC functions is that it takes an input stream and
> breaks it up into a number of bands, each of which has a fixed data width.
> The first thing that happens is that the psychoaccoustic model is used to
> determine how many bands or channels are allocated to different frequency
> ranges. That is, if a chunk of music has a lot of high frequency sounds
> but few low frequency sounds, more bands will be allocated to the high
> frequency ranges and fewer to the low frequency ranges. More bands means
> greater resolution and fidelity. Once bands have been allocated, each band
> is processed with the psychoaccoustic model to remove sounds that cannot be
> heard.
Not even close. There are no "bands, each of which has a fixed data
width", unless you're calling a bit a band.
Ignoring Short Block Mode for a moment, each 11.6ms of audio (one
soundgroup; 512 samples @ 44.1khz) is transformed into the frequency
domain, yielding 512 coefficients (numbers). The overall goal is to
allocate some number of encoding bits to each of those 512
coefficients such that any inaccuracies in their representation
(called quantization error) will be hidden inaudibly below the masking
threshold (which is a property of human psychoacoustics revealing that
we don't hear nearly as much as we imagine we do). Note that there are
only 1696 bits (212bytes/soundgroup) available for storing those 512
values, hence an average of 3.3 bits per coefficient if we wanted to
store them all at the same resolution. Fortunately much of the musical
signal is below the masking threshold, hence we can get quite good
resolution for the parts that count using only a fraction of the
original amount of data.
There *are* a set of fixed (frequency based) groupings of these 512
spectral coefficients, within which the coefficients share a common
scaling factor and mantissa length.
> Improvements to ATRAC are really improvements to the psychoaccoustic models
> used for encoding. The "format" of data on a disc is identical; the
> difference is in which bits are recorded and which are not. And as the
> psychoaccoustic model has absolutely nothing to do with playback, there is
> 100% forward and backward compatability.
This part is right.
Rick
-----------------------------------------------------------------
To stop getting this list send a message containing just the word
"unsubscribe" to [EMAIL PROTECTED]
