[Audio-Me]

Source: Sony Classical SK 57499 DDD recording SBM (Super Bit Mapping 20-bit master)
Ax•Stoltzman•Ma - Brahms•Beethoven•Mozart Trios for Piano, Clarinet & Cello
Track 5: Beethoven - Trio Piano, Clarinet and Cello in B-flat Major Op. 11; Allegro con brio 9:10
original wav size: 92.5MB
------------------------------------------------------------------------------

LAME - *.mp3
Version: 3.92.

Command line options:
Constant Bit Rate
-b320 -q0 -k -ms
-b320 320kbps bitrate
-q0 use highest quality algorithm
-k full bandwidth, disable all filters
-ms normal stereo mode

noticeable difference from original wav
output file: 20.9MB aprox. 4:1 compression ratio

Variable Bit Rate
-V0 -q0 -k -ms
-V0 highest quality vbr

more noticeable
output file: 154kbps average bitrate 10.1MB aprox. 9:1 compression ratio
------------------------------------------------------------------------------

MPC musepack (MPEGplus) - *.mpc
Version: 0.90s. Optimized by Frank Klemm

Command line options: all variable bit rate
--insane --nmt 99 --tmn 99 --ms 0
--insane highest quality preset
--nmt/tmn nmt is min, tmn is max, increasing number upto 99 results in higher bitrates
--ms 0 normal stereo mode

unnoticeable
output file: 584kbps average bitrate 38.2MB aprox. 2:1 compression ratio
lol, bigger file than lpac

--insane -nmt 16 --tmn 32 -ms 0

barely noticeable
output file: 323kbps average bitrate 21.1MB aprox. 4:1 compression ratio:

--insane -ms 0

much less noticeable than mp3
output file: 257kbps average bitrate 16.8MB aprox. 6:1 compression ratio
------------------------------------------------------------------------------

LPAC - *.pac
Version: 1.40.

Command line options: -0 -i
-0 custom compression
-i normal stereo mode

lossless
output file: 33.6MB aprox. 3:1 compression ratio
------------------------------------------------------------------------------

MAC - *.ape
Version: 3.96.

Command line options:
extra high compression mode

lossless
output file: 31.8MB aprox. 3:1 compression ratio

------------------------------------------------------------------------------
*footnote: the pre-compiled versions I have linked are all win32, you may find them for different operating systems from their sites.

 

[dngl]

thanks for the comparison A&M. Hopefully the bootleggers will switch to LPAC...

 

[Audio-Me]

I just found out about command line version of Monkey's Audio. Its got better compression.

But, its got delayed seek, as in random access isn't as quick as LPAC (which needs to be enabled in the encoding).

 

[yage]

Interesting... though, why didn't you use mid/side coding with LPAC?

 

[JohnV]

I'm not trying to defend Lame, but your Lame settings are not the highest quality.
Firstable -q0 uses new kind of noiseshaping (noiseshaping has many meanings, in this I mean the mp3's noiseshaping in quantization loop) which is not working correctly in 3.92. There's a serious quality affecting bug according to the Lame dev Takehiro Tominaga. So you shoud use just -h (or -q2).

Try --alt-preset insane for highest quality 320kbps cbr
and --alt-preset extreme for highest quality VBR.

These modes use mostly and most of the times stereo-frames, although to increase quality mid/side frames are used when it can be done safely.

Also I recommend the use of latest MPC encoder, mppenc 1.02.
http://www.uni-jena.de/~pfk/mpp/bin/...ndows-1.02.zip
0.90s is pretty old. 0.92s introduced a new and improved tonality estimation (According to the MPC dev Frank Klemm, old encoder with tmn 18 is less tonal than new with tmn 16, and there's better bass tonality estimation in general).

 

[Joe Bloggs]

Hey JohnV long time no see! Greetings from Jo at r3mix.net forums!

They still haven't fixed q0 huh? A&M if you included compression time--I bet it took next to forever using -q0 too

 

[Audio-Me]

MP3 still sounds the same with those settings. I'll give the newer MPC version a try. I've always hated the use of joint stereo.

encoding time is irrelevant

 

[JohnV]

To me it sounds different, better most of the times. Very rarely worse.

Joint-stereo is a controversial issue. There's joint-stereo and joint-stereo. The thing which matters is the implementation and thresholds how often mid/side frames are used. Most of the hate comes from FhG implementation.
Lame's joint-stereo is tweakable. I've seen people "hating" the JS, even though all the frames would be stereo-coded. Of course then it's only psychological...

Btw. nice to see you like MPC. I was very much involved in MPC's listening testing and tweaking development phase up to Andree's 1.7.9. Still of course checking quality pretty often.

I wouldn't recommed the use of stereo coding with MPC below insane profile. If you have high channel correlation and you use stereo coding, it may lead to significant quality reduction. This is because of the effect called stereo unmasking, and because of the fact that MPC currently doesn't do BMLD (Binaural Masking Level depression) calculations when using stereo coding.. The use of mid/side coding prevents quality decrease in this case. Also MPC's joint stereo can be controlled in subband (32 subbands) level, making it very much more quality oriented.

 

[Joe Bloggs]

Us people from the music compression forums are experts at telling whether something is pschological or real...

To further confirm your results, A&M, you ought to running a comparison program like ABX... and see if you can still hear differences from the original wav in a blind test...

 

[Audio-Me]

I'd rather not let a program tell me what I'm hearing.

I did blind tests for my results (eyes shut, random). I also got four samples from atheren for a blind test (all decoded to wav) and I could not discern difference of original and mpc encoded with the xtreme preset (or so he says, hehe). My test was with familiar music, the other was with music I've never heard before.

the other two samples were mp3 and original with lowpass filter (some reason this one sounded weird to me, yet not the mpc encoded one which also has hf cut-off)

 

[Ricky]

Quote:

Originally posted by Audio&Me I'd rather not let a program tell me what I'm hearing.

It won't tell you what you're hearing, it will just make sure the tests are double-blind, and analyze your results statistically.

Double-blind methods such as ABX or ABC/HR are the unique accepted methods by the developer/scientific community to rate perceptual encoders. In fact they are the actual methods used to develope these psychoacoustic compression schemes.

 

[Joe Bloggs]

Problem is, it has to be run on a computer, and it is not easy to build a good audio system around a computer.

Audio&Me has such a system, however