Compress MP3 and MP3Pro by 20% losslessly? Options

[jetpower]

Hi,
I did not know if to create a new topic or just post in one of mp3packer threads but I figured it has to be important.

Somewhere on the Internet I found information about this software called SoundSlimmer. They claim the following:

Current version of SoundSlimmer losslessly compresses Mp3 and Mp3Pro files averaging 20% less file size, while general purpose archivers are capable to reduce file to 1-2%.

and how? this is what they say

SoundSlimmer works with audio data only utilizing certain specifics of digital sound signals to losslessly slim the audio content.
---and---
The core of the SoundSlimmer is a semantic approach to the data that represented music and correlation analysis of spectral data patterns called sound genes .

http://www.soundslimmer.com/
You can download the software there, too. The software can be downloaded as 30-day trial (valid e-mail needed).

-
I tried it with a few files. On first three *.mp3s I got an average 10% reduction in size, but on the forth (one hour-long speech 56kbps) mp3 file it exceeded 20% . The newly created files got an *.mpz extension. So their claim doesn't seem that far off.
I did not check if the software decodes properly, however.

So, is it really a revolution in mp3 compression?

[goodnews]

Looks promising.

If it works as promised, and they say future versions are planned to work for AAC, AACPlus and MPEG 4 audio, this should be a winner and affect the future of compressed audio in a substantial way. Maybe MPEG and/or Coding Technologies would want to license this company's patents/code for inclusion in a future enhanced version of MP3, AAC, AACPlus, MPEG 4 audio, etc.

Since they say their software will put it back to bit for bit identical of the original MP3 also, this should be interesting to test if anyone has the time and tools for it. I hope someone either proves this works or exposes it for not working. Interesting theory - sound genomes...

[skamp]

Infima 2: Return of the Scam?

[jetpower]

Infima 2: Return of the Scam?

Did you test it? I don't think it can be compared to Infima. For example it's only a 600kb file.

[skamp]

It also requires you to give your email address before you can download their program. Thanks, but no thanks.

[Shade[ST]]

It also requires you to give your email address before you can download their program. Thanks, but no thanks.

Tried giving 2 email addresses 20 minutes ago, still no program (or URL) in my inbox.

[guruboolez]

I tried, and it seems to work.
6 files were compressed; gain=15%
Then I decompressed the mpz files, and compared the result with the original: bit-to-bit identical. It's definitely not comparable to the Infima April joke.

What I didn't checked is if the app is working like mafuka and if it stores the spared data somewhere on the HDD

Anyway, if this process would be interesting on small flash players, I fear that it's near useless otherwise (or maybe for filesharing?).



EDIT:

Wow:
http://audiotests.free.fr/temp/soundslimer.png

Original files were 8kbps "ghost" LAME encodings
(I'm using them for tagging purpose on a second computer)

This post has been edited by guruboolez: Apr 29 2006, 00:00

[goodnews]

' date='Apr 28 2006, 04:48 PM' post='387364']

It also requires you to give your email address before you can download their program. Thanks, but no thanks.

Tried giving 2 email addresses 20 minutes ago, still no program (or URL) in my inbox.

Look in your spam filter, that is where GMAIL put my E-mail from them with the download link and authorization/unlock registration code. Got the E-mail in a minute or so after signing up on their site, but those nasty spam filters in Gmail caught it.

[Thundik81]

CODE

SoundSlimmer development team:

                      Algorithms: D. Vatolin

Dmitry S. Vatolin : http://graphics.cs.msu.su/en/people/index.html
Real lossless compression, real improvement

Paper (in russian) : http://www.compression.ru/dv/course/compr_audio.pdf

This post has been edited by Thundik81: Apr 29 2006, 00:22

[Shade[ST]]

What I didn't checked is if the app is working like mafuka and if it stores the spared data somewhere on the HDD

Try moving the archives to another computer and decompressing them. Then, move the original files, and bit-compare. You might want to try with different content, too. (eg. a real mp3)

[Demetris]

I run a test on 10 files encoded by LAME 3.97b2 -V 0 --vbr-new
After decompression, MD5 hashes match.

CODE

Test system:          Win XP Pro SP2, AMD Athlon 64 3000+
Program version:      SoundSlimer 1.04.001

Total time:           1:08:24
Compression time:     0:09:07                           [ 7,5 ]
Decompression time:   0:08:21                           [ 8,2 ]

Original size:        122.742.332 bytes
Encoded size:         107.246.467 bytes                 [ 87% ]

FILES

Kind hearted woman blues (Take 1) (R. Johnson)          [ 84% ]
La vie antérieure (H. Duparc - B. Kruysen & N. Lee)     [ 86% ]
Prélude No. 17 Ab - Allegretto (F. Chopin - M. Pollini) [ 86% ]
Sister Ray (The Velvet Underground)                     [ 86% ]
Tom the model (Beth Gibbons & Rustin Man)               [ 86% ]
Vénus callipyge (G. Brassens)                           [ 87% ]
Allegro energico... (G. Mahler, Sym. No. 6 - C. Abbado) [ 88% ]
Milk it (Nirvana)                                       [ 88% ]
Dancing queen (ABBA)                                    [ 89% ]
Squirrel song (Shellac)                                 [ 89% ]

The program is 602kB and installs cleanly (monitored with Total Uninstall 2.35).

Here is a quote from the latest issue of Squeeze Chart (Edition 2006 04 02) by Stephan Busch.
(Download here: http://www.maximumcompression.com/benchmarks/benchmarks.php)

Sound Slimmer 1.04.001 (06.07.2005) - a very very cool lossless! recompressor for MP3 and mp3PRO audio files, capable of reducing every MP3 song from 12 % to 46% without loosing quality; without cheating; without deleting; reaching new world record on MP3 testset, and all MP3 files I compressed with this demo available at www.soundslimmer.com could be easily extracted bit by bit; even checksums matched.

Despite that, Sound Slimmer handled MP3 files created with all known encoders on the market, accepted mono, stereo and joint stereo formats at all imaginable CBR bitrates and VBR encoded titles.

MP3's containing newer music are usually reduced by 12%, music from the 80's by 20% and most classic or silent relaxation titles up to 36 to 46%.

[haregoo]

What I didn't checked is if the app is working like mafuka and if it stores the spared data somewhere on the HDD

Try moving the archives to another computer and decompressing them. Then, move the original files, and bit-compare. You might want to try with different content, too. (eg. a real mp3)

Completely matched on separate machine.

It's looks promising.

Edit:As I think it as lossless compression like 7z, coding speed is relatively slow in my opinion.
but compression performance is very good.

This post has been edited by haregoo: Apr 29 2006, 04:04

[chrisgeleven]

I want to know how they did this.

[guruboolez]

' date='Apr 29 2006, 01:21 AM' post='387389']
Try moving the archives to another computer and decompressing them.

I can't do it for the moment. I searched on the whole HDD for all files created in the last 24 hours, and the tool didn't return any bad result. No, this tool is really amazing. I tried with others MP3 files (-V5 encodings) and the gain was always ~15%. For three hours of music at ~130 kbps, I spared 20 MB. No bad

[Garf]

I want to know how they did this.

The linked paper is in Russian, so I can't make sense of it, and it looks like a general overview of existing things anyway.

But I can make some very educated guesses. It probably works in the same way as the lossless JPEG recompressors, i.e.:

1) Replace huffman coding in MP3/MP3Pro with arithmetic coding
2) Replace the scalefactors which are independent in any frame by versions that are differentially coded in time. (Perhaps try the same for M/S data, code groups, etc)
3) Use predicition on MDCT data.

The "problem" of all these techniques is the increased decoding complexity (compare how fast you can decompress an mpz to how fast you can decode an MP3).

[hlloyge]

More that that, I want that incorporated in lame encoder by default

H.

(edit) Oh, I see, it would be very complex to decode these streams... but it could be useful as an option.
I know it isn't free, but this is just wishful thinking...

This post has been edited by hlloyge: Apr 29 2006, 11:10

[seanyseansean]

More that that, I want that incorporated in lame encoder by default

H.

But it wouldn't be an mp3 anymore.

[guruboolez]

Including that in LAME would break the compatibility with thousands of hardware and software decoders.

[SirGrey]

The linked paper is in Russian, so I can't make sense of it, and it looks like a general overview of existing things anyway.

This paper contains a general information on audio coding shemes for younger students, no more.
Something I didn't hear about is hybrid coding - an attempt to use wavelets, FT and noise shaping simultaneously to get better results on low bitrates, but as I understand paper, this tehnique failed.

[hlloyge]

Including that in LAME would break the compatibility with thousands of hardware and software decoders.

I noticed I wrote nonsense I had a blonde moment.

H.

[Sebastian Mares]

I wouldn't like to see this in LAME, I would like to see it in PowerArchiver, WinRAR and/or 7-Zip.

[Phobos]

it looks very promising imho...

for a PC is a no brainer, and i believe most Symbian Compatible Celullar phones might already have enough horsepower to decode mpz.

The biggest downfall is its popularity, look at mpc... it has still not a lot of hardware support and has been around for years. I agree with goodnews that it should be licensed by the big players like mp3 or aac to have a chance in the market.

Anyway it might take years to become popular.

I know i wrote a lot of "mights", but thats the way things have been behaving lately in the audio scene

[smok3]

whats are the decoding times?

[rutra80]

Look what for example WinRK, StuffIt, or PAQ can do with JPEGs.
I wonder if soundslimer replaces noisless techniques used in MP3 encoding with its own ones (optimized more for compression ratio than speed) as Garf said, or maybe it is just optimized for squeezing the last bits from huffman-encoded bitstreams. Can someone try to rename some OggVorbis or AAC file to .mp3 and try to feed soundslimer with it?

This post has been edited by rutra80: Apr 30 2006, 05:08

[haregoo]

Can someone try to rename some OggVorbis or AAC file to .mp3 and try to feed soundslimer with it?

Vorbis : It did not work.
LC-AAC : Compression ratio was about 98%.

But if it works well with AAC, Apple will be interested in this technology for saving bandwidth which iTMS consume.