How to spot fake FLACs

[Nguyen Trung Hoa]

How do you tell whether the source of a FLAC file is lossless or a lossy (I call the latter a travesty)?

Do you want to be 100% sure? Then grab one of your CDs and rip it to FLAC. It's a lame answer to a lame question, but there really is no other way to be 100%, as you will see. I guess this is one of the perks of buying the CDs.

But you can use spectral analysis to get a pretty good idea whether the source is lossless or lossy, so you can sleep all warm and fuzzy at night :)

The tool of choice is Adobe Auditon. I used v1.5 (I'm an old school guy), but any more recent version will do just fine.

I picked one of my original CDs (Metallica - Live ~Oh Deficate!~: Binge & Purge CD1 if you want to know) and ripped one track to a .wav file using EAC. I then encoded the .wav file to each of these formats (highest bit rate possible):
1. MP3: LAME MP3 320kbps CBR
2. AAC: NERO AAC 1.3.0.0 400kbps CBR
3. Musepack: MPC 1.30.0 q10 (~350kbps)
4. Ogg Vorbis: aoTuVb5.7 mod 500kbps CBR

After encoding to each of these formats, I decoded to a .wav file to generate the spectral with Audition. It's as simple as dragging the .wav files to the Audition's windows (choose the "Edit View" tab first), and then View -> Spectral View. You can find a link to a pack with all the pictures at the end of the post (the second one).

Original

This is what the spectral looks like for the original file. A lossless source looks smooth (no blocks) with no frequency gaps. These are the two fingerprints of lossy codecs. A low cutoff, say at 18Khz, is suspicious. But without blocks or gaps, it is more likely that the microphones used to record the sound weren't able to capture higher frequencies.




MP3

You can imediatelly sense that something is wrong with this one. Above 16Khz you can see lots of gaps and dark spots. A trained eye flags this as lossy imediatelly




AAC

Also easy to see something isn't right. The 18khz cutoff is very sharp, and above 18Khz we have some gaps.




MPC

Impossible to call based on this image. Looks virtually the same as the original (and don't forget, in real life you won't have the true lossless file to compare it to).




Ogg

Idem.




Let's take a closer look, to see if we can identify the MPC and Ogg as lossy. I like 2 second zooms. The first images for MPC and Ogg didn't give us a clue where to look at. So I'll have to choose some random section.

Original 2s




MPC 2s

You can see a difference now. Look at the higher frequencies in the 46.6-46.7s and 46.8-47.0s intervals (among other places). MPC removed them. Subtle, but the difference is there. But could you spot call it without the original to compare it to? I don't think so.




Ogg 2s

This one is even closer. Barely any diffence. Look at 47.3-47.35s interval.




MP3 2s

This one is a piece of cake. Those flying bricks (blocks with gaps) don't lie. This is what we're looking for (or not).




AAC 2s

The sharp cutoff gives it away. Notice those gaps, and square edges. But no flying bricks.




So, MPC and Ogg are giving us a hard time. But the differences are there. And likely they are more pronounced at some other places. The problem is that we don't know where to look at. Or do we? The problem here is that the difference between no signal (black) and the faint signal (dark blue) the lossy codec removes, is difficult to see... with these color scheme. So, let's change it. Go to options -> settings -> Colors tab -> Spectral Tab, select "Spectral High Energy" on the list, and "Change Color..." to black, and repeat for "Spectral Low Energy" and choose white. Finally, set the "Gamma" to 0.4. What we get may not be as pretty, but is way more effective.


Original B&W




MPC B&W

Can you see the difference now? You could call it now. The mics are recording up to 21khz, but still there are places where there is basically nothing (noise, no matter how faint) there between 19khz and 21 khz. Very suspicious, but I'd like to see some blocks




Ogg B&W

Idem.




MP3 B&W

A walk in the park.




AAC B&W

Like stealing a candy from a baby (hey, you shouldn't do that).




Let's look at the 2 second zooms to see some blocks (the zoom is at the same place as before)

Original B&W 2s

Can you see any blocks? I don't think so.




MPC B&W 2s

Razor cut blocks. That's more like it. No flying bricks, though :( But it can produce those bricks, if you look at the right place on the right track.




Ogg B&W 2s

We might not have sand castles, but some bricks feel like getting airbourn (47.05-47.1s interval, top channel)




MP3 B&W 2s

Looks ugly doesn't it?




AAC B&W 2s

Easy to tell the bricks.




Let's sum it up:
- MP3 is very easy to spot.
- AAC idem, but with a caveat. There are several codecs and profiles. Some might be way better and thus harder to spot
- Ogg Vorbis is a tough nut to crack. But this is at 500kbps. At VBR q9 (~320kbps) it produces many more blocks, and it's much easier to spot (but still much harder than MP3 or AAC).
- MPC is another tough nut and, worse, at 350kbps. This is a plausible bitrate to be used, so it's plausible that someone encoded with MPC at 350kbps and someone else transcoded to FLAC, resulting in a difficult to spot fake.

Is this it? Unfortunately no. You can use Adobe Audition to add noise, and merge it with the lossy source. This would darken the bright white areas. The blocks would still be there, but it would be harder to know where to look at (see the next example to see what I mean) and spot the blocks. The blocks wouldn't be so evident either. I have no doubts that, with some lipstick, you can make an extremely high quality Ogg or MPC file look like lossless. But an important thing to keep in mind when judging whether it is lossless or not, is to think if it is reasonable to be lossy. If it looks so much like a lossless, there are two possibilities: (1) it is lossless; (2) it is an extremely high bitrate (which is rare) of a not so commonly used codec (compared to MP3) with noise added to make it pass as lossless. (2) is possible, but not plausible. The fakes you see are from posers who grab the first MP3 they find and transcode it to FLAC just to... well, pose.

Continues...