seanadams wrote: > One thing to note when doing this kind of math: for lossless formats > (whether FLAC or WMA doesn't matter), as you go from 16 bits to 24 > bits, you can expect your compression ratio to go down substantially. > That is to say, you will get almost no compression to speak of on the > next 8 bits, because they are mostly noise.
Clearly if the lower 8 bits are Gaussian noise, its not going to compress well. It won't compress period. It seems to me that the impact of this depends a lot on the level of compression of the music. If the music has been mangled/destroyed by Loudness Wars, then most of the signal will be close to full scale, with at least 4 and probably 6 bits of noise at least. If there is any dymanic range left, for a lot of the signal, the upper eight, ten, and even 12 bits will be zero, which compress well. I've seen no evidence that humans need more than 96dB or so of real signal to noise, which the RedBook spec allows in theory. So 18 or 20 bits is really enough for humans, add a couple more bits for processing and bad dithering. Until there are a fair number of real 24 bit recordings, its all speculation, but I'd love to do some analysis on what the real signal is on commercially available 24 bit recordings. -- Pat Farrell PRC recording studio http://www.pfarrell.com/PRC _______________________________________________ audiophiles mailing list audiophiles@lists.slimdevices.com http://lists.slimdevices.com/lists/listinfo/audiophiles
