Thanks to all. This is compiled from lists archive, but I think that in archive is not oldest messages.
So. MP3 is not useable as archive quality at any bitrtate. Because of frame overlapping, impulse response is distributed over longer time than one granule, typically two (I am right?). Regards J. Lukesh ------- Frank Klemm says: MP3: length of granule [ms] fs block [kHz] long short - ------------------------------------- 8 72 24 11 52.2 17.4 12 48 16 16 36 12 22 26.1 8.7 24 24 8 32 18 6 44 13 4.3 48 12 4 For critical music granules down to 2 ms are useful, but not available in MP3. :: > Listen to castagnets at 32 kHz/320 kbps and at 48 kHz/192 kbps. :: > The 48 kHz/192 kbps sounds much better. This is a restriction of MP3. :: :: ??? What is the basic reason for that ? MDCT window size. It is 192 or 576 in samples. This is a little bit too long in time for 44.1 and 48 kHz and much too long for 32 kHz. AAC uses 128 or 1024, which is nice for most music. AAC 48 kHz/128: 2,7 ms AAC 44 kHz/128: 2,9 ms --------- "Ivan Dimkovic" <[EMAIL PROTECTED]> says: Short window length is 192 samples (this is the most important reason) With this block length, MP3 has impulse response of 18.6 ms at 48 kHz sampling rate. Human perceptual system has more advanced psychoacoustic in time domain (less than 5 ms in some cases). MPEG commitee had chosen shorter window length for next generation of coders (AAC, MPEG-4 General Audio, ..) -------- Segher Boessenkool <[EMAIL PROTECTED]> says: ----- The optimum window size for strong changes is about 128 samples (44 or 48 kHz), because all pre-echo will be inaudible. The optimum window size for stationary signals is infinite. > The AR function inside a granule should be constant within 3...5 dB. > This can be for >100 ms for music like some parts of "Adagio for Strings", > but also 0.8...1.2 ms for synthetic percussion. _______________________________________________ mp3encoder mailing list [EMAIL PROTECTED] http://minnie.tuhs.org/mailman/listinfo/mp3encoder
