IPB

Welcome Guest ( Log In | Register )

Multiple sequential reencodes as test for lossy formats/codecs (MP3,Vo, with links to .cmd script and .cmd batch converter/tagger
metafizik
post Feb 2 2012, 20:05
Post #1





Group: Members
Posts: 1
Joined: 2-February 12
Member No.: 96891



Do not know why, but today I wrote .cmd script program that performs multiple sequential reencodes on the same audio file. It gave out some interesting results and somewhat undermined my inclination to Vorbis aoTuV vs AAC.

At this moment I have tested only on single audio file, which is soundtrack from CRPG, close to classical music genre, without voice, and have compared lame MP3 (-q0 --abr 156), Vorbis OGG aoTuV v2.87 (-b 170), Nero AAC (-q 0.476) by audible quality and spectograms of resulting audio. File sizes were about equal for all three formats after 1st encode.

Some conclusions from this experience (which is very limited for today):

1) MP3 format is the worst. Quality degrades quickly and signal levels (loudness) lower significantly. File sizes lower too (from 3.4M at 1st encode to 2.9M at 60th). 50th encode is absolutely unbearable. And lame.exe -q0 MP3 encoder is significantly slower than oggenc2.exe and neroaacenc.exe

2) Vorbis aoTuV is better than MP3, but quality degrade significantly faster than with Nero AAC. Opposite to MP3, Vorbis obviously increases signal levels, although the gain is not too audiblе (but evident in spectrograms colors).

3) Nero AAC is the best of three. Even 100th encode is still relatively bearable, at any case much better than 50th MP3.

All three codecs show quality degradation in common, similarly sounding "effect". Do not know adequate English term, something like wavy hissing... maybe someone is able to call it with right word. For Vorbis OGG the effect is most apparent; it seems that almost all of Vorbis distortions come to this effect; it sounds very consistently, wavy and monotone. MP3 has prove itself to be worse than Vorbis because the simular effect sounded much more inconsistent and accompanied by very unpleasant artifacts/glitches, especially on "difficult places" of audio track. AAC also had the considered effect, but clearly distinguishable only after dozens of reencodes; no noticeable glitches, spectrograms "lower" but do not become as bad as for MP3 and Vorbis.

=============

Maybe there are people on this forum who have no difficulties with command line and parameters of CLI encoders and who want to test this method of multiple reencodes for different kinds of music and for other codecs? In this case there are two archives. The first contains .cmd scripts and some small helper .exe:

http://f.rodon.org/m2/audcvt.rar

Script audlossy.cmd performs reencodes by calling another more complex script audcvt.cmd - batch audio converter/tagger for different formats.

To run sequence of encodes just place source file (audtest.flac, for example) into any directory and run script in that directory:

C:\WORKDIR>start audlossy.cmd 30 10 audtest.flac aac "-q 0.47"

it must cycle through 30 reencodes and save each 10th result, thus creating numbered .aac files for 1st, 10th, 20th and 30th encode.

When started, audlossy.cmd automatically detects the last of all saved files for the given name so it is always possible to continue reencodes further.

audcvt.cmd converter supports some other lossless and lossy encoders (CT AAC, Fraunhofer AAC, Musepack, ffmpeg wma, Speex; see --help; more can be added relatively easily by ones with knowledge of .cmd scripts). audcvt.cmd lists all required .exe in its help (when run without arguments). Archive with all required executables has the same URL as path in previous link with different file name: audcvt-external-exe.7z

audcvt.cmd can generate spectogram pictures in batch mode, e.g. "audcvt ogg png" creates .ogg.png spectrograms for all *.ogg files in current directory (by converting each of them to .wav and using sox.exe)

=============

Multiple sequential reencodes method brings up some questions. How universal is the above experience from single track? How much it depends from genre, bitrate, codec (for given format)? Can we consider rate of quality degradation as objective measure for audio format/codec comparision? Or maybe some algorithms, like AAC, just have special property of not lowering quality significantly only in case of repeated reencodes, being not so good on other types of input? What AAC encoder is "the best": Nero, CT, FhG, QT? Audio is not my biggest interest, but maybe someone is interested enough to examine these things, or maybe there are even people who already performed similar tests or have knowledge on this subject and can share it?

Go to the top of the page
+Quote Post

Posts in this topic


Reply to this topicStart new topic
1 User(s) are reading this topic (1 Guests and 0 Anonymous Users)
0 Members:

 



RSS Lo-Fi Version Time is now: 16th April 2014 - 09:35