OK, who brought up this rumour that MPC isn't good enough at low bitrates, the Vorbis mafia?
No, harpsichord lovers
I really like this instrument, and noticed that harpsichord blows the bitrate up with musepack. At low setting, bitrate is really high, and quality significantly low. Curiously, the same encoding with --alt-preset are normal (but quality bad :-/ )
mpc 1.14 --quality 5 = 226 kb/s [standard]
mpc 1.14 --quality 4 = 182 kb/s [radio]
mpc 1.14 --quality 3 = 134 kb/s [thumb]
mpc 1.14 --quality 2 = 94 kb/s [telephone]
mpc 1.14 --quality 1.99 = 63 kb/s [below telephone]
lame 3.90.2 --alt-preset standard = 187 kb/s
lame 3.90.2 --alt-preset extreme = 223 kb/s
Lame VBR extreme is smaller than mpc --standard !
The track is from J.S. Bach, Fugue BWV 892 in B major, 1994, recorded by Pierre Hantaï. All harpsichord tracks are turning round the same bitrate. This is a common situation for harpsichord recordings.
In comparison, the ogg table for the same track :
ogg vorbis 1.0 -q-1 = 49 kb/s [+ 9%]
ogg vorbis 1.0 -q 0 = 64 kb/s [0%]
ogg vorbis 1.0 -q 1 = 79 kb/s [-1%]
ogg vorbis 1.0 -q 2 = 99 kb/s [+ 3%]
ogg vorbis 1.0 -q 3 = 135 kb/s [+ 20%]
ogg vorbis 1.0 -q 4 = 163 kb/s [+ 27%]
ogg vorbis 1.0 -q 5 = 204 kb/s [+ 27%]
ogg vorbis 1.0 -q 6 = 237 kb/s [+ 23%]
Vorbis and mpc have the same reaction : bitrate explosion. LAME sucks. --alt-preset standard sound really bad (at least on the first notes, pure and sharp harpsichord), and I'm able to abx preset-extreme & preset-insane too (but quality is OK). It seems that psycho-accoustic model of mp3 (or just lame) is incomplete, buggy, or prehistoric (don't know).
At low bitrate now, mpc gives the most awfull sound of all encoding. I can't describe the 64 kb/s mpc file, even in french . Telephone is awfull too, vaguely the same than --alt-preset 64 [lame 3.93.1 test version] - with more chirps for mpc, much larger (+50%). Sound is muffled in the two cases.
Vorbis, in comparison, is fantastic. No care about snowy sound in that situation : distorsion is high (higher than mp3 encoding at 64 the same bitrate), but the harpsichord sound sharper, 'clearer' (much treble).
Acceptable sound is reached by ogg -q 3 [135 kb/s] and --radio profile for mpc [195 kb/s].
==> modem encoding are not conceivable with mpc for some music. If --radio profile gives a really good quality (at least for me), the quality drop under this profile begin to be worrying. The small comparison I did in the past between different codecs never surprised me and went (for once ) in the same way than common opinion. And I often find mp3 to be a more reasonnable choice than mpc (muffled sound, but low distorsion level). SBR may be helpful, but I'm not sure that time should be spend for tweaking mpc at such bitrate.
Has Dibrom or someone else a real explication (not supposition) for the unability of lame to properly encode harpsichord, by underrating the bitrate necessities ? Is this a bug, or an inevitable flaw for mp3 format ?