Hi Everyone,
I don't wanna interfere into the discussion whether or not ATRAC encoding is
possible on todays PCs (please no flames!!), but.... I have an ATRAC
encoder/decoder running on my PentiumII-450Mhz (with good old Win95b). I named my
encoder/decoder ATRAC Type-WB (guess why!) Version 3.5-0.91 (this means the
psycho-acoustic model has been read out from a MD-Deck with ATRAC-Version 3.5 --
MDS-JA3ES).
The encoder eats about 75-80% of the CPU time and the decoder needs 20-25%. Both
encoder and decoder are written in a high-lever language and have yet not been
opimized (no MMX and so on). The encoder even writes the bitstream in the exact
way a MD recorder would do, which means, if you write the encoded file onto a
MiniDisc, set the pointers in the UTOC correcly and put the disc into a MD player
it will play. About 15% of the CPU time is used for the QMF and the type
conversion (float2integer) -- unbelievable; 10% is needed for the bit
packing/unpacking and the fast cosine transforms. A lot of improvement could be
done here since I use the FCT algorithm for the DCT type I, but ATRAC uses DCT
type IV. Anyway. My encoder version (0.91) has still some problems with the
BlockSize decision.
Since sony has registered the name "ATRAC", I am already thinking about a new
name for my encoder/decoder. How about "CARTA" (pronounced "ATRAC" -- if you read
it from right to left! :) Any suggestion are welcome! BTW: Everybody with some
math knowlege can build their own en/decoder. All the information you need is
availble on the net (no kidding!)
wolfgang
ps: I now know why Sony announce their new gear (MDS-JB940) so early. Well, I
somehow have to prepare my parents about what my wishes for next chistmas are!
Although, if I remember correctly, my Mom told me if I ever come home with new MD
stuff, I will be kicked out of the house (that I wouldn't really care... well
yes, I would) and she will throw away all my MD stuff (that I surely WOULD
care!!)
----------------------------------------------------------------------
Stainless Steel Rat wrote:
> * Eric Woudenberg <[EMAIL PROTECTED]> on Sat, 11 Mar 2000
> | I'm missing something. Didn't I demonstrate that MP3 (a more complex
> | coder than ATRAC)
>
> Whosoever has told you that is full of crap. At the fundamental level, MP3
> (MPEG-1 Layer III audio) has a pathetic time-frequency distribution model.
> At lower bitrates, 128Kbps (allegedly "near-CD quality) and below, audio
> signals with lots of high and low frequency sounds but relatively little in
> the middle get... "cropped" is the best way I can describe it. When the
> time-frequency block is full, any remaining signal gets thrown away,
> whether or not it is significant. I understand that Prince's "Raspberry
> Beret" is an "MP3 breaker" for this reason, but I have never made the
> comparison personally.
>
> ATRAC has a much more sophisticated time-frequency distribution model, one
> that uses the same psychoaccoustic model as the bit reduction algorithm.
> The result is that time-frequency blocks are allocated based on "density".
> Frequency ranges with more sound get more bandwidth; those with less sound
> get proportionally less bandwidth.
>
> | runs at 3X realtime on my 500MHZ desktop machine? Do you still somehow
> | think ATRAC would be slower than realtime on such a machine?
>
> For sound quality as close to ATRAC 4 as MP3 is capable of achieving, yes.
-----------------------------------------------------------------
To stop getting this list send a message containing just the word
"unsubscribe" to [EMAIL PROTECTED]
