Lame APS nukes MPC 5 on Amnesia.wav, What the #@? Options

[Pio2001]

Here are my listening tests results.
Parameters -quality5 --xlevel

Amnesia

Amnesia44.wav vs Amnesia44-114.mpc : ABX 8/8
Amnesia44.wav vs Amnesia44-115.mpc : ABX 8/8
Amnesia44-114.mpc vs Amnesia44-115.mpc : ABX 8/8

Amnesia48.wav vs Amnesia48-114.mpc : ABX 8/8
Amnesia48.wav vs Amnesia48-115.mpc : ABX 8/8
Amnesia48-114.mpc vs Amnesia48-115.mpc : ABX 8/8

Here's my subjective quality ranking :

Amnesia44-114 : 0/5
Amnesia44-115 : 2/5
Amnesia48-114 : 4/5
Amnesia48-115 : 4.5/5
Original : 5/5

Conclusions :
The 48 kHz sound better than the 44.1 kHz (no ABX)
The mppenc 1.15 alpha sounds better than the 1.14 (improvement ABXed on both samples)
The bad 48 kHz sounds a bit better than the good 44.1 kHz IMHO (no ABX)

Transwave

Transwave.wav vs Transwave-114.mpc : ABX 8/8
Transwave.wav vs Transwave-115.mpc : ABX 6/8. Again : 7/8. Total 13/16, probability I was guessing 2 %
Transwave-114.mpc vs Transwave-115.mpc : ABX 4/8

Subjective ranking :

Transwave-114.mpc : 4/5
Transwave-115.mpc 4.5/5
Original : 5/5

Conclusions :
1.15 alpha seems better than 1.14 again, but this sample is too weak for me to ABX the two versions of Mppenc.

[Case]

Pio2001's transwave sample.

[Dibrom]

Here is a new version from Frank which may improve upon these issues some more:

This version was greatly modified to better recognize signals which you can
found in amnesia, 2nd_vent, jump, beck, awe32_20sec, angelic, ...

Code is programmed, but parameters are not fine tuned very well (I had only
tuned 2 days long). More information about tuning you can find in
mppenc.info in the source package. When you want you can play with these
parameters ...

Tuning not only includes quality, but also avoiding bitrate waste.
CD quality is at least at 1411,2 kbps possible which much easier means ;-)

Compilability under Windows is not tested do to lack of working MS VS.

Remarks are welcomed. Flames should be encapsulated by <flame> </flame>.

mppenc.info

win32 binaries (courtesy of Case - thanks )
linux binaries

[Frank Klemm]

Here is a new version from Frank which may improve upon these issues some more:

This version was greatly modified to better recognize signals which you can
found in amnesia, 2nd_vent, jump, beck, awe32_20sec, angelic, ...

Code is programmed, but parameters are not fine tuned very well (I had only
tuned 2 days long). More information about tuning you can find in
mppenc.info in the source package. When you want you can play with these
parameters ...

Tuning not only includes quality, but also avoiding bitrate waste.
CD quality is at least at 1411,2 kbps possible which much easier means ;-)

Compilability under Windows is not tested do to lack of working MS VS.

Remarks are welcomed. Flames should be encapsulated by <flame> </flame>.

mppenc.info

win32 binaries (courtesy of Case - thanks )
linux binaries

Note that that version is for testing, because it is a post SV7 pre SV8 bitstream version.
When the results are positive I port the result back to SV7.

[Tim Mervielde]

Here are my WinABX-results for some “problem-samples” encoded with mpcenc1.95z67 (default settings).

-------------------------------------
WinABX v0.23 test report
03/10/2003 02:46:57

A file: H:\AUDIO TEST SAMPLES\mppenc1.95z67\angelic_silent_short.wav
B file: H:\AUDIO TEST SAMPLES\mppenc1.95z67\angelic_silent_short.mpc.wav

02:48:12 0/1 p=100.0%
etc…
03:04:59 41/64 p= 1.6%
03:05:04 test finished

-------------------------------------
WinABX v0.23 test report
03/10/2003 03:06:12

A file: H:\AUDIO TEST SAMPLES\mppenc1.95z67\amnesia44khz.wav
B file: H:\AUDIO TEST SAMPLES\mppenc1.95z67\amnesia44khz.mpc.wav

03:06:31 1/1 p=50.0%
03:06:45 2/2 p=25.0%
03:06:57 3/3 p=12.5%
03:07:16 4/4 p= 6.2%
03:08:13 5/5 p= 3.1%
03:08:34 6/6 p= 1.6%
03:09:22 6/7 p= 6.2%
03:09:49 7/8 p= 3.5%
03:09:55 test finished

-------------------------------------
WinABX v0.23 test report
03/10/2003 03:09:57

A file: H:\AUDIO TEST SAMPLES\mppenc1.95z67nd_vent_clip.wav
B file: H:\AUDIO TEST SAMPLES\mppenc1.95z67nd_vent_clip.mpc.wav

03:10:54 1/1 p=50.0%
etc…
3:16:18 12/16 p= 3.8%
03:16:25 test finished

-------------------------------------
WinABX v0.23 test report
03/10/2003 03:21:51

A file: H:\AUDIO TEST SAMPLES\mppenc1.95z67\Jump.wav
B file: H:\AUDIO TEST SAMPLES\mppenc1.95z67\Jump.mpc.wav

03:23:06 0/1 p=100.0%
etc…
03:32:01 15/21 p= 3.9%
03:32:05 test finished


-------------------------------------
WinABX v0.23 test report
03/10/2003 03:36:54

A file: H:\AUDIO TEST SAMPLES\mppenc1.95z67\AE_BASSCADET.wav
B file: H:\AUDIO TEST SAMPLES\mppenc1.95z67\AE_BASSCADET.mpc.wav

03:37:45 0/1 p=100.0%
…
03:46:24 19/38 p=56.4%
03:46:26 test finished

****

It should however be noted, although I could abx 4 out of 5 samples (not easy!), that the improvement in sound quality of the samples was impressive, compared to mppenc v1.14 and 1.15 encoded samples. I would rate them all higher than 4.6 on a scale of 5.

There is however a serious problem with the blips-sample (Blips.flac)

Watch this

Encode log:
__________

MPC Encoder 1.95z67 --Alpha-- © 1999-2002 Buschmann/Case/Klemm/Piecha

encoding file 'H:\AUDIO TEST SAMPLES\mppenc1.95z67\blips.wav' to file 'H:\AUDIO TEST SAMPLES\mppenc1.95z67\blips.mpc'

SV 15.15, Profile 'Standard'

PCM fader : fade-in: 0.00 s, fade-out: 0.00 s, shape: 1
Scaling input by : left 1.00000, right: 1.00000
Maximum encoded bandwidth: 22.0 kHz
Adaptive Noise Shaping : max. 6th order
Clear Voice Detection : Dual
Mid/Side Stereo : Mid/Side Stereo when superior + enhanced (2/6 dB)
Threshold of Hearing : Model: min(Busch,Klemm) (5), Max ATH: 76 dB, Offset:
+0 dB, +Offset@20kHz:0 dB
Noise masks Tone Ratio : 6.5 dB
Tone masks Noise Ratio : 18.0 dB
Exploitation of temporal post masking
Minimum Signal-to-Mask : 1.0 dB
No deleting of input file after encoding

%|avg.bitrate| speed|play time (proc/tot)| CPU time (proc/tot)| ETA
100.0 5981.6 kbps 4.21x 0:11.3 0:11.3 0:02.6 0:02.6

_______________
An average bitrate of nearly 6000kbps!

But then there’s more….

Decode log:
_____________
MPC Decoder 1.95z67 3DNow/SSE © 1999-2003 Buschmann/Klemm/Piecha/Wolf

decoding of file 'H:\AUDIO TEST SAMPLES\mppenc1.95z67\blips.mpc.mpc' to file 'H:\AUDIO TEST SAMPLES\mppenc1.95z67\blips.mpc.wav'
5968.2 kbps, 0:11.36, SV 15.15, Profile Unstable/Experimental (--Alpha-- 1.9)

0:00.02/ 0:11.36 decoded ( 0.0%)

mppdec: broken frame 66/435 (decoded size=3691, size in stream=3232)

(runtime: 0.23 s speed: 7.45x)

*** 1 decoded file has errors ***

_____________

The encoded blips.mpc would not decode, and the decoding aborts with a broken-frame error.

-tm

[Pio2001]

Is it an effect of my imagination, or is there a treble dropout on the right channel at 10.0 s in this sample : http://perso.numericable.fr/laguill2/files...s/kovenant.flac ?

I couldn't ABX it successfully, but I was not far :

Mppenc 1.15r -quality5 --xlevel : ABX 6/8
Mppenc 1.14 -quality5 --xlevel : ABX 6/8

SoundForge shows that it's clipping... It can explain the (possible) difference, but it's the first time I hear clipping as a drop out.

[KikeG]

As a global, 6/8 + 6/8 = 12/16, p=3,8%... very likely there's a difference.

[NumLOCK]

Have you tried the new settings ( http://static.hydrogenaudio.org/extra/mppenc.info ) ?

[niktheblak]

Is it an effect of my imagination, or is there a treble dropout on the right channel at 10.0 s in this sample

Cool, there's nothing like ABX'ing a great song like Mirror's Paradise

I have that album encoded with Andree's old mppenc, using -xtreme setting. I will do a few ABX trials with a recent mppenc this evening.