auCDtect - is a program for determining the authenticity of musical CD records, using the character of the audio information it contains, and predict its possible source is it an original studio record or a reconstruction from a lossy encoded data (for example, MP3).

Current version: 0.7
Differences from 0.6 version:
- Improved algorithms.
- More stable dithered mp3-s detection
- Shows probability original/mp3 instead of former Unknown area.
- Improved accuracy.
- Tested on more than 7.000 musical files.

Linux version
Windows version
Algorithm details

Best Regards,
Oleg Berngardt, auCDtect developer

Great util. One question, the new build guesses the wav as MPEG or CDDA
with a probability of %, so when it reads a decoded (ex:) mp3 extreme as
"This track looks like CDDA with probability 82%". What exactly does that mean.?
Would that by any chance refer to how well a encoder have done.?
(am using it with questionman.)

Thanks for the updated version.

I have made a windows shell integration to scan a whole directory.
Maybe you can use it

auCDtect.reg

Great to final see this type of program. I'm not aware of others of this type?

FireStarter,
Basically, this means, that:
1) Program is not absolutely sure about source of this file (when it sure, it should be 100% probability) - such properties have both originals and mp3s.
2) According to our statistics, 82% of original tracks have similar properties and 18% of mp3s, so in most cases it should be CDDA...
3) If you would send me this mp3 to aucd@true-audio.com and describe procedure of obtaining .wav, i would be happy to include it in our database...
Best Regards,
Oleg.

WaldoMoster,
Thank you!

Nice tool really..Works nice in context with Mrquestionman

nice

a MP3 file (Lame 3.91 preset Extreme)
conversion with foobar2000

• with dither:
  looks like CDDA with probability 71%
  
• without dither:
  looks like MPEG with probability 100%

julien_32,
What kind of noice shaping do you use? Or just dithering with none noice shaping?
Best Regards,
Oleg.

• no dither:
  looks like MPEG with probability 100%
  
• no noise shaping
  looks like MPEG with probability 100%
  
• traingular noise shaping:
  looks like MPEG with probability 99%
  
• strong ATH:
  looks like CDDA with probability 73%
  
• soft ATH:
  looks like CDDA with probability 63%

another test:
the same file with same conversion properties

• first time:
  looks like CDDA with probability 36%
  
• second time:
  looks like CDDA with probability 40%
  
• third time:
  looks like CDDA with probability 73%

!!!!!

It's a good results, why not? ;-)

As you can see - noise shaping masks MPEG artefacts.

auCDtect 0.6 could not check files after dithering;
auCDtect 0.7 can to check files even after the noise shaping except the ATH.

Thus, I think, really it is not required to detect such files.

---
Alexander

Bern think about publishing source code (maybe some lightened version) so i (or someone else) could
-make altlinux rpm for official sysiphus
-integrate it with kaudiocreator/kscd/other software

Shafff,
Unfortunately, for now we do not have opened code of auCDtect
But RPM is now available too at ftp://true-audio.com/aucdtect-0.7b-1.i386.rpm .
It is build/tested under RedHat-7.2, so it should be installed correctly under AltLinux too...
Tomorrow we will check it under new SuSE too...
Thank you for good suggestion!
Best Regards,
Oleg

updated debian packages are already available in the RW Debian Rep.


...later

and have you been interested in wavelets?

as for me, i'm only 17 and i'm entering IT department of university only this year. we must have wavelets in our programme, but i'm not sure when... :-(

Bern:

Just a heartfelt thanks from this non-techie end user for this immensely helpful program, and I look forward to the ongoing development with great interest.

auCDtect fails completely on lowpassed recordings of any kind (says they're 100% mpeg even if it's a true audio cd that was lowpassed at 20kHz upon production for whatever reason)

And of course it gives pretty random results when analyzing analog radio / tv broadcast recordings, but that's to be expected I guess..

Unfortunately, we do not use any of wavelets in our program, we use a bit different algorithms

ssamadhi97,
We tryning to use very old records during testing too, for example Shaljapin (it is russian singer, begining of 20 century, 3KHz bandwidth), and it works well. About any records with cut above 20KHz made by any program or technique - this creates sound artefacts close to mp3-artefacts and because of this detected as mp3s. Thank you very much for note, we will extend our collection by additional number of old records...