Author Topic: Bitrate checkers  (Read 1558 times)

hiccup

  • Hero Member
  • *****
  • Posts: 6562
In a blind test, without seeing the graph beforehand, would you really be able to the difference between same track at 320 mp3 versus FLAC? I have nothing to back this up, but I would guess that most people wouldn't notice a difference.
Depending on the quality of the equipment, the recording (acoustic ambience etc.), and (probably most important) 'knowing what to listen to', it's not so hard to identify lossy vs. lossless.
But for most people, for most recordings, for most equipment it will be extremely difficult if not impossible indeed.

hiccup

  • Hero Member
  • *****
  • Posts: 6562
Dropping all of the flacs on the batch file and seeing if any look decimated in the upper frequency ranges like the spectrograms in this post would be an efficient way to accomplish this.
This is a lossy mp3:



Would you have guessed looking at the spectrum?
 

frankz

  • Hero Member
  • *****
  • Posts: 3317
Obviously not.  I guess I'm a fool with a bunch of fake flacs in his collection. 
A smile is happiness you'll find right under your nose.

hiccup

  • Hero Member
  • *****
  • Posts: 6562
Obviously not.
The one thing that gives it away is if you look the first ten seconds. There at 16kHz you see a step in the density.
That's something you'll almost always see with mp3's.
Opus files don't have these 'steps', but Opus has a hard frequency limit at 20kHz, which you will often notice if you compare it against the lossless version.

vincent kars

  • Sr. Member
  • ****
  • Posts: 380
if the file is completely lossless I should see frequencies present all the way up to 22Khz
Can be the case but it depends very much on the recording.

The highest possible frequency a PCM audio file can contain is half the sample rate (Shannon/Nyquist)  
So in case of CD it is 44.1/2=22.05
To avoid any frequency above 1/2 fs, a filter is used. A filter always has a slope so needs some "space"
In practice you will see frequencies up to 21 kHz and then a very steep drop.


However, this is what is technically possible, not to be mistaken for the properties of the recording.


This is rip of one of my CD's. It is a string septet.  Indeed lossless and no musical life above 15 kHz.
Almost all chamber music on CDs form the 80's, 90's behave this way.

audiobabble

  • Jr. Member
  • **
  • Posts: 24
Obviously not.
The one thing that gives it away is if you look the first ten seconds. There at 16kHz you see a step in the density.
That's something you'll almost always see with mp3's.
Opus files don't have these 'steps', but Opus has a hard frequency limit at 20kHz, which you will often notice if you compare it against the lossless version.

I'd have to disagree with that, surely the step in density is to do with what is happening in the music. It's very rare that you will find a track that starts at full volume from the get-go, so most music will have this noticeable change in density regardless of what format it is in.

hiccup

  • Hero Member
  • *****
  • Posts: 6562
I'd have to disagree with that, surely the step in density is to do with what is happening in the music. It's very rare that you will find a track that starts at full volume from the get-go, so most music will have this noticeable change in density regardless of what format it is in.
Well, you are wrong.
Here is the same track as Opus encodes it:



The density bump at a specific frequency is a typical signature of mp3. (at least for Lame it is)
 

BoraBora

  • Jr. Member
  • **
  • Posts: 30
Ultimately, I prefer to trust my ears and can easily pick out a 320 mp3 next to a flac or wav file, although could be (and have been) fooled if all I'm hearing are 320 mp3's
It's almost impossible to hear differences between a properly encoded 320 kbps MP3 and its lossless version. You can occasionnaly identify such an MP3 by its pre-echo artefacts in some instruments like castanets, but not much else. I'm talking about a controlled double-blind test, like an ABX, which is the only test you can trust (for yourself or other people).

Even with golden ears and good training in spotting MP3 artefacts, there is no way you can "easily" pick out a 320 kbps from a lossless. So here are some leads:

* You're comparing two different editions with mastering differences
* One of these files has been converted from a sub-par version
* The two files have different sound levels (always use Replaygain before comparing)
* Placebo effect

Please don't feel insulted by the last one. Placebo effect is human and all of us are subject to it, without exception. ;)

As for verifying if a lossless album hasn't been converted, there's a very trustful way in most cases (not all): AccurateRip. This can be done on a complete ripped CD only (be it a single or a full album), but not lone tracks. Download and install CUETools: http://cue.tools/wiki/CUETools_Download#Download . In the Action box, check Verify. Then drag and drop your album folder (with or without a .cue) in the Input box and press Go. A log will be displayed in the main window. Set the options to your needs (writing the log in the album folder, recursive analysis etc.).

The log will give you results from both the CUETools and AccurateRip database. The most trustful is the AccurateRip one. Most logs will look like these ones (I edited them down to the important parts) :

Quote
[CUETools log; Date: 05/01/2016 00:16:18; Version: 2.1.6]
[AccurateRip ID: 000826fb-003728c0-71068c08] disk not present in database.

This is a CD-R home-produced by friends, never released to anyone but our circle. Obviously, it's missing from the AR database. But what if it was a commercial CD? Then you have to worry about the validity of this supposed lossless rip. No ID in the AR database can only happen if nobody ever ripped this CD in an AR-featured ripper (like MusicBee, obviously, but also Foobar2000, XLD, dBPowerAmp and lots of others). This doesn't mean this rip is not perfectly fine and 100% lossless but there's a doubt. If it's a well-known album, then it can be a transcoding or a counterfeited indonesian edition etc.

Quote
[CUETools log; Date: 16/01/2022 19:01:16; Version: 2.1.6]
Pregap length 00:00:33.

[AccurateRip ID: 0010e797-0092d206-89095a0b] found.
Track   [  CRC   |   V2   ] Status
 01     [d59f6fd8|19db8a5d] (0+0/3) No match
 02     [19487148|207beb90] (0+0/3) No match
 03     [084c706c|e2cad204] (0+0/3) No match
 04     [8fcbba20|92e24646] (0+0/3) No match
 05     [404bf14b|0638f188] (0+0/3) No match
 06     [4a0c4d05|42e2884c] (0+0/3) No match
 07     [d244f8bb|b25f4117] (0+0/3) No match
 08     [7bc34679|e04ecb07] (0+0/3) No match
 09     [1865b616|7e0e92ec] (0+0/3) No match
 10     [bcf3ed5a|78e44ac6] (0+0/3) No match
 11     [a221e222|ca0f5756] (0+0/3) No match
Offsetted by -6:
 01     [8f5321ca] (0/3) No match (V2 was not tested)
 02     [72840ea0] (0/3) No match (V2 was not tested)
 03     [12bfccfc] (0/3) No match (V2 was not tested)
 04     [c87b82f6] (0/3) No match (V2 was not tested)
 05     [65fa96b5] (0/3) No match (V2 was not tested)
 06     [67882f7b] (0/3) No match (V2 was not tested)
 07     [cd0ae745] (0/3) No match (V2 was not tested)
 08     [6c6cb3f9] (0/3) No match (V2 was not tested)
 09     [29213c7e] (0/3) No match (V2 was not tested)
 10     [6956ce94] (0/3) No match (V2 was not tested)
 11     [ee6dbdca] (0/3) No match (V2 was not tested)

This CD has an AR ID with 3 copies registered but the rip doesn't match any of the 3. It may be suspicious or not, depending on the rarity/popularity of the CD. In this example, this is an obscure 1998 blues album from John Brim and Pinetop Perkins, released by an austrian label. One one hand, my copy could be from a reissue. Very small labels don't press huge quantities and they often repress on demand. On the other hand, Discogs lists only one edition of this CD. But maybe this rip hasn't been made in a secure ripper and isn't bit-perfect. It doesn't mean it's a transcoding but there is a doubt.  

Quote
[CUETools log; Date: 13/01/2022 23:37:49; Version: 2.1.6]

[AccurateRip ID: 00148b2f-00dc767c-c309930e] found.
Track   [  CRC   |   V2   ] Status
 01     [a6af0a10|37d4a747] (10+23/36) Accurately ripped
 02     [0e12503b|2000c2de] (10+23/36) Accurately ripped
 03     [86b307e6|f913b7d7] (11+24/38) Accurately ripped
 04     [e30a6e9f|ba7067a9] (10+24/37) Accurately ripped
 05     [f4fceca6|88d1b62e] (10+24/37) Accurately ripped
 06     [d75cf4cb|6dc314a1] (10+24/37) Accurately ripped
 07     [9230d059|7c65d3f7] (10+24/37) Accurately ripped
 08     [37ffdacb|76631d5e] (10+24/36) Accurately ripped
 09     [104d2651|2c3e10e1] (10+24/37) Accurately ripped
 10     [fae45614|934b713a] (10+24/37) Accurately ripped
 11     [b1d3552e|1e6dc70f] (10+23/35) Accurately ripped
 12     [d4bec96a|50446e9e] (10+23/36) Accurately ripped
 13     [b995b13a|1223f04a] (10+24/37) Accurately ripped
 14     [e4bb4c10|d5e5bf6a] (10+24/37) Accurately ripped
Offsetted by -1540:
 01     [1915db1e] (03/36) Accurately ripped
 02     [4f4b6e8a] (03/36) Accurately ripped
 03     [c17095f8] (03/38) Accurately ripped
 04     [7e01b51b] (03/37) Accurately ripped
 05     [5ee22f08] (00/37) No match (V2 was not tested)
 06     [b911904a] (03/37) Accurately ripped
 07     [04fc996a] (03/37) Accurately ripped
 08     [b8e8fa7d] (02/36) Accurately ripped
 09     [de757d6a] (03/37) Accurately ripped
 10     [7f47797d] (00/37) No match (V2 was not tested)
 11     [fede457f] (02/35) Accurately ripped
 12     [3aee5f85] (03/36) Accurately ripped
 13     [85e9e706] (03/37) Accurately ripped
 14     [566f4672] (03/37) Accurately ripped        
This one is from the 1989 edition of The Complete Capitol Recordings, Vol. 1 of Art Tatum. No doubt here: this is the real deal. More than 30 people have ripped this CD and got the exact same CRC. You don't have to worry about a possible transcoding. This is the kind of log you'll see most of the times. Even 1 or 2 matches means you don't have to worry. No transcoded rip can match a logged lossless rip in the AR database.

Quote
[CUETools log; Date: 16/03/2022 17:44:38; Version: 2.1.6]
Padded some input files to a frame boundary.
[AccurateRip ID: 0046f85a-052d641a-7412071a] disk not present in database.

No doubt either with this one, but in a bad sense. The "Padded..." line means it's been transcoded. Not all transcodings can be identified by this padding, but if CUETools had to pad each track to a frame boundary, then you're 100% sure there was a transcoding in the chain.

As you can see, this CUETools verification isn't just about transcodings but also about bit-perfect rips, so it's inducing an additional level of paranoia and despair.  ;D  After all, is a non-transcoded but badly ripped CD still lossless? You also have to know some commercial CD can be mastered from MP3. They're rare but they exist, especially with labels reissuing public domain music. Lots of bad stuff here, soundwise. They don't have access to the tapes so they source their CD from whatever they find. But some well-established labels did this, too.

Again: it's rare, no need to worry much about it, just be aware of it. I have such a CD, a expanded reissue of an english 80's pop band, The Maisonettes. The label is Cherry Red, a perfectly legitimate english label, the music is copyrighted, and nonetheless the CD was mastered from MP3. Not a big deal, it's not a great album or a great band, I bought it for memories and I'm not even sure it would make a difference if it was mastered from tapes.
Last Edit: March 16, 2022, 06:13:35 PM by BoraBora