Mp3 War

This text is a small summary of a text I made for personal use on early July 98 using my home language. Since this text was more than 6 descriptive pages, I’ll try to include as much as I can. Pls do not link, copy, post part or all text without permission. Michael Hondrosmos (mymailbox@iname.com)

1) Encoders Categories
2) The "Cost" of Compression
3) The Results
4) Today ?

  All mp3 compressors are divided to 3 main categories:

1. Fraunhofer Group: fundamental company that helped a lot for the mp3 development. It also makes hardware encoders.

Encoders:

i) L3Enc 2.72: Fraunhofer’s Dos Encoder (the only pure Dos encoder) . Due to the difficult (command line) user interface, there are many Front-Ends for Windows (i.e. L3Enc Win Manager) .

ii) L3Enc 3.0: Recent 32 bit version. It’s not compatible with previous version command line set.

iii) Fraunhofer libraries (dll): Fraunhofer .dll files . When installed they can be detected from Settings -> Control Panel -> Multimedia -> Advanced -> Audio Compression. All applications using these libraries make exactly the same .mp3 files (with the same characteristics). The only thing they change is GUI. You can find them searching the Internet for l3codec.zip or l3codecp.exe . These libraries does not support more than 128 kbps.

iiia) OptiCom mp3 Producer (professional): Windows application. The Pro versions supports more than 128 Kbps .

iiib) Mp3 Compressor 0.9f : Pirate program -illegally- using these .dll’s . You can find it by searching for «Mp3Compr09f» . It supports command line, but it cannot change its bitrate (from command line).

iiic) The rest: CodeAxe etc: All the other programs using these libraries .

iiid) Audio Grabber etc: CD DA programs (CD audio extraction) using libraries to save directly to mp3. However it is better to extract to Wav and then encode it, rather than extract directly to mp3, because CD-Rom drive works much more time (the speed degrades from 20x to 0.2x ).

 

2. Xing Group: Popular company for its audio/video products, cards etc. Developed its own algorithms and projects for .mp3 .

Encoders:

i) Xing Mpeg Compressor 3.20: Windows application. Uses MMX technology to speed up process. It uses command line (but it is undocumented).

ii) Tompg: Primal Xing version. Dos text mode but works under Win95. You can find it searching for «tompg» . Although a lot older than Xing 3.2 , the mp3 resulting files are exactly the same.

iii) The Rest: Cosmo Mpx Encoder, Mpeg Creator, YunaSoft Encoder, Mplifier 0.54 ?.a. The common pint is the use of Xing application. Most of them work with Tompg and other versions - even unregistered - of Xing’s Mpeg encoders. All of them produce exactly the same file type.

iv) Xing Mp3 Compressor 1.00: The newest Xing product. With obvious audio improvements, retrieving less CPU resources (with respect to ToMpg & Xing Mpeg Compr.), Its command line is L3enc compatible (using x3enc.exe) . H win application does not support all bitrate rate (i.e. 160 Kbs) but the Dos command line (x3enc) does. The only FAST program that we can run it in background with no noticeable interruptions..

3. ISO coders: All the other programmers, teams that tried to realise the mp3’s ISO standard (which is freeware and can be found at Fraunhofer site) .

The most popular encoders are:

i) Mpeg Encoder 0.07 (SoloH). The first totally legal free encoder. One from the oldest encoders. Used by many people as point of reference because of its high quality compression. It does not support command line.

ii) BladeEnc 0.54. Dos text mode but works under Win95 and supports Drag ’n Drop. Also supports L3enc command line emulation. The program is freeware and it’s source code is given to Cd-Copy 4.504 (so CD-Copy is the first CD Grabber with integrated mp3 encoder. It does not need any libraries or external program). The shareware version is full .

iii) 8hz Encoder 1.1b: Primal version. You can find source code of previous versions in its site. Supports command line .

iv) The Rest: Make-it-Mp3 etc: There are so many out there and they are growing fast ...!

The Cost of Compression

It is obvious that the only mp3 - or any other compression - disadvantage is the loss of «information» (apart from the fact that we are loosing the audio-cd format so we cannot play songs to typical cd players).

From time to time we see many benchmarks (i.e. www.mp3bench.com) but you cannot find anywhere a research for mp3’s detailed «behavior». So this is the reason of my research.

The thesis of these tests is the following:

Since we know that we will have loss of information, we will examine how .mp3 «sounds» like , not how «looks» like under an oscilloscope with a pink , white -or whatever- noise, square or triangles waves etc.. All tests are audio acoustic comparing results without any «laboratory examination». If you disagree upon these thesis then do not read the rest.

My «tools» consists of:

1) Pentium 233 MMX 64MB Ram.

2) Sound Cards tested: Opti 931, SB Awe32 Asp, SB 16, SB 64 OEM, SB 64 Gold .

3) Hi-Fi with: Rega Planar 3 / RB-300, Stanton 681, CD Marantz 63 MKII KIS, Amp Marantz PM80 MKII , Audio Spectrum - Artemis (Greek floor based hi-fi speakers using SEAS cones).

For the test purpose I encoded many , many samples . Since my project was to transform all my Vinyl & CD collection to mp3, I extracted songs from more than 400 vinyl records or Cds. So I had the opportunity to listen to many samples. From all this experience I separated the samples I found some differences with the originals.

Finally I kept 3 .wav which I used as pilot samples to make further experiments:
Wav1 : Digitally recorded trechno-rave song with high freq drums .
Wav2 : Vinyl recorded jazz song with long tailed physical drums (snares, cymbals).
Wav3: Vinyl recorded jazz song with long tailed female vocals .

Generally the sound of an 128 kbps mp3 file can be heard pleasantly. But if we focus with great attention to sound weaknesses we will lead to 3 major categories:

1. Hi-Freq Cut. Most of encoders using a software cut-off. If the cut-off freq is low (16 Khz) or the program is not well constructed , it results to a muddy, blurred trebles. Anyway this is the smallest weakness .

2. Trebles phasing: Mostly Fraunhofer encoders results strange distortion in high frequencies. For example I observed that songs which have many long physical drums (snares or cymbals) the trebles sounds as they ‘ve been processed by a phaser or such device: Continuous phase and volume increases - decreases . The phenomenon is more audible in vinyl recordings. Generally this phenomenon is present mostly to encoders using Joint Stereo . So avoid - if possible - Joint Stereo mode and prefer (pure) Stereo. All tests below made with pure Stereo mode (when I had the ability to choose)

3. Vocals phasing: Mostly Xing encoders results strange distortion in upper-middle frequencies. Songs that have long tailed vocals without any other instruments (female or other hi-freq), the vocals vibrates and we can notice a small whistle between vocal maneuvers , «s»’s or phonemes. This phenomenon is also noticeable in other cases like: solo acoustic or electric-jazz guitar, wind instruments (trumpet , sax etc) .

Ironically, most of 3^rd category encoders (3^rd party ISO), that includes the ISO standard without any patents or variations does not present such phenomena very noticeable. The main reason is that the two mentioned companies (Fraunhofer / Xing) are oriented mostly for low bitrated applications (Internet Radio, Video/Audio) where we demand low throughput. Encodings below 56Kbs have much more better results using the above mentioned companies products.

The Results

As mentioned above I used 3 wav (one for each phenomenon). I did not took under consideration program's GUI (how user-friendly it is) , but only how well it bypass all mentioned phenomena and how fast do it (plus time variable). All variables are equally weighted. So Grad1 are my results. In case you are not care about time Grad2 are the results without time element.

Pat.: Patent type (Fraunhofer, Fraunhofer dll, Xing, 3^rd ISO )
Time: Test1.wav (28 sec) on P/233 MMX under Win95
Speed: (10=fast, 1=slow) [ Sqrt( Sec(min) / Sec(i) ) x 10 ]
C1: Hi-freq clarity (10=unnoticed, 1=distortion)
C2: Trebles Phasing (10=unnoticed, 1=distortion)
C3: Vocal Phasing (10=unnoticed, 1=distortion)
Grad1: Average grade
Grad2: Average grade without Time variable.

I. 128 Kbs

- All other programs belong to the 1^st or 2^nd encoders category, the sound behavior (C1, C2, C3) will be the same as the «Fr.dll» / «Xing» programs. The only difference is GUI and elapsed time .

If we equally weight all variables the best encoder is Xing’s Tompg. If we do not care about time then the best is BladeEnc 0,54 . Finally in case we work only with vocals (or similar sounds) the optimal choice is Mp3 Compressor 0.9f (or L3Enc 2.71 as dos equivalent).

The comparison table lead us to one conclusion: There is not perfect encoder. If you looking for something better increase bitrate. The differences over the 3 categories seems to be equalized and minimized above 160 Kbs.

II. 160 Kbs

In this bitrate level .mp3’s are about 24% bigger so the compression rate follows down to 1:9 (instead of 1:11 @ 128 kbps). Unfortunately Fraunhofer dll (with all supported programs) do not allow compression greater than 128Kbs. All categories tends to extinguish their weakness. Again here, BladeEnc 0,54 presents less faults than the rest and Tompg remains the fastest best choice.

(*) 8hz : Starts with a pop in the beginning of the .mp3

III. 190 Kbs (not translated . similar results)

Generally encoder’s weakness are more audible in vinyl (or analogue recorded) 60’s, classic rock, jazz or similar sounds (with many solo vocals, snares and wind instruments). If you want to encode something like that prefer 160 kbps or more.

Out of List Programs:

1. Sound Limit 1.5 : ISO program. Failed both C2 & C3. GUI but not friendly (does not save previous used paths or settings), does not support command line.
2. Ampeg 4.3: Dos ISO encoder. File format incompatibility and few available settings.
3. All other programs based on Fraun. libraries or Xing engine.
4. Make-it-Mp3:. Failed to C1 , C2 , C3
5. MpegDJEncoder 1.0: Integrated solution from Xaudio for CD-to-Mp3. However it is an «amateur» application without many required tools & settings needed .
6. Music Match Jukebox 2.0: Similar Cd-to-Mp3 solution. Anyway I think that audio extraction and mp3 management are complex tasks that cannot be solved with programs like that.
7. L3enc 3.0: . Does not seem to have great differences with it’s previous versions.

04/07/98 - Michael Hondrosmos - mymailbox@iname.com

P.S. What happens today.

As you can see from the program version numbers 6 months is a long time for a computer science research !!. I made this test primarily to observe compression weaknesses. You can use these experiments to make your tests one year after or with any newer encoder.

Maybe this research is a crap because there more weaknesses not detected. But anyway this is an answer what I mean «test» . I ‘m tired hearing words like «quality» , «lousy» without describing what they did or even what they mean. Anyway whoever talks about «quality» or «loss» must give his wav & mp3 (cause & effect) to everybody, to see what he means .

Regards
Michael H.