Oh, boy, I know these recommendations very well, I've been using LAME in CLI mode only since several years and tried many presets, -q options and other "improvements". The best what you can set is: --CBR -b 320 -q 0, and this is what I referred to.
I know the alghorytm of JS very well, it's a mathematically lossless operation and gives 100% replication of data when used with lossless
compression. When you take lossy compression, Mid and Side signals are compressed lossy
which means that M and S signals could cause some data loss while L and R signals replication. Have you ever thought about the fact that Side has a very specific nature and it should use a different psychoacoustic alghoritm to compress it?
|And you're mistaken about joint-stereo. It doesn't destroy spatial data, simply, if the same data exists in both channels, it makes it unnecessary to have two complete copies of it.
I mistook nothing. I'm writing about my aural impressions, not speculating about what could have happened to the data. Again - Joint Stereo is lossless only if the whole compression is lossless. It's due to different compression errors applied to M and S signal compression. FhG seems to have better alghoritms as I prefer their sonic fidelity to both Stereo and Joint Stereo versions of LAME.
I suggest you listen more than read Hydrogen Audio. I really appreciate open-source codecs and think the ogg Vorbis by aoTuV is the best existing lossy codec (comparable or better than AAC) but LAME - sorry, there is too much of self-praise, probably due to placing mathematics above psychoacoustics which was the origin of the mp3 invention.