True FLAC vs. Fake FLAC Options

[Goratrix]

You get a match against a verified yet errant submission from a download and it doesn't matter how many submissions from legitimate sources are made.

Yes, I don't deny that. But it's a question of statistics and trust. What I've been doing for the last year or so is a little test. For every CD that I bought (around 50 albums in 2010, a mix of old and new releases) I also downloaded a lossless rip (before someone goes "pirate!", that's actually legal where I live). I ripped my original and compared it to my downloaded copy, and looked at the AR results, also in a few months time again, after AR got populated, if it was a new album. The result was that ALL of those downloaded rips were genuine (same as my original CD), except two, and even those could have been a US vs. EU pressing/mastering difference and not a fake.

Is that a conclusive scientific test? No it isn't. But it's enough for me to trust that statistically, the majority of illegal lossless rips are in fact geniune.

It's the same dilemma as when doing a rip yourself, you never know if it's 100% "perfect". But statistically, most of them are, so why worry about the 1% or 5% (or whatever) that aren't? If someone is so worried, they should buy the original, as always.

[XeR0]

Also, when you think about it, the OP's question really doesn't make sense. The result of a FLAC compression is a lossless image of the source. In case you input a mp3-compressed audio, does that make the resulting FLAC "non-lossless" or "fake"? No, it doesn't. It makes it a perfect lossless version of the source material, regardless of what the source material is :-)))))

On another note, if we are talking about identifing "fake FLAC" in files obtained by...ehm, questionable means, then AccurateRip comes in quite handy

Technically speaking, you are correct. And while my original question may not make any sense, I did explain what I meant by True FLAC vs. Fake FLAC. MP3's discard audio data. You can decode it into WAV and then compress it into FLAC. My question was, how would I be able to know the difference between a FLAC file derived from an MP3 file vs. a FLAC file derived directly from WAV which was derived from a CD. But I've already gotten an answer to that question so I guess there's no need to continue that particular discussion any further.

You get a match against a verified yet errant submission from a download and it doesn't matter how many submissions from legitimate sources are made.

You'll probably get a couple of matches but assuming the album is popular, the amount of mismatches against correct rips would be larger.

[greynol]

Assuming you're using a tool configured to show you mismatches and exercise ample skepticism, sure.

[ChrisC7]

Looking at a spectrogram of the audio is a good way to tell, but you need to look for more than just a low-pass filter.

Here are three spectrograms made with SoX that show what to look for:

First, the lossless original:

http://farm6.static.flickr.com/5160/588503...19f4fd5_z_d.jpg

Next, Lame MP3 at -V5:

http://farm7.static.flickr.com/6023/588503...51ec1c8_z_d.jpg

And lastly, Vorbis at -Q4:

http://farm7.static.flickr.com/6011/588503...a8664a0_z_d.jpg

The main thing I look for is the "holes" in the audio that the lossy compressors psycoacoustic model determined to be inaudible due to auditory masking.

Rotareneg,

Thank you for the info. Can you please tell me how to make spectrograms like that? I went to get the SoX program but I see that it's all command-line and I'm not very good with that sort of thing. I use the Spectro program (http://spectro.enpts.com/) which is great and very easy to use by right-clicking a FLAC or mp3 and selecting "Analyze," but the spectrograms are very small and I'd like to make big ones like you did. Could you please tell me what you type at the command-line to make those? Thank you very much.

This post has been edited by db1989: Aug 7 2011, 14:34

Reason for edit: demoting inline images to links

[Porcus]

The only only to really know (to my knowledge) if a CD is authentic is to buy it from the publisher themselves.

I guess Century Media doesn't qualify as a publisher?

Seems to me that this should be an entry for the knowledgebase wiki.

unless you've ripped the files yourself, how would you know the FLAC file you have is ACTUALLY lossless instead of an MP3 converted into FLAC?

As others have pointed out to you, there is no bullet-proof way to tell whether you have gotten the source, because the source sold by the record company may have been mp3'ed along the way. Possibly even from the band themselves. There are ways to verify with fair confidence that the material has been through lossy processing though (falsifying this could possibly be harder, I don't know), but without a «better» source to compare with, there are basically no ways to tell that what you got, was not the «least lossy» copy there is.

I've used the TEST option in FLAC frontend and it doesn't give a result. I have used Audiotester and it does say the file failed because it's TRUNCATED.

That sounds like an encoder/decoder issue.

[astroidmist]

There used to be a freeware DOS command line/console program that could examine a WAV and tell if it came from an MP3. I tried it out on my personl files and it was NEVER WRONG even though I couldn't tell by listening and needed to double check the sources. Unfortunately I can't remember what this program was called or where to find it. It ran on Windows 98 SE too. I am really bummed that I can't find it.

[Northpack]

There used to be a freeware DOS command line/console program that could examine a WAV and tell if it came from an MP3.

It's called AuCDtect and does a spectrum analysis looking for patters introduced by lossy compression. It works very well, you'll hardly ever get a false negative. There a windows frontend called Tau Analyzer and even a foobar plugin, all avaiable here: http://en.true-audio.com

[dreamliner77]

The only only to really know (to my knowledge) if a CD is authentic is to buy it from the publisher themselves.

I guess Century Media doesn't qualify as a publisher?

What I meant was this: Let's say for instance I wanted to buy an album called "Tales of the Inexpressible". A trustworthy publisher would be "Twisted Records" instead of "World Music of New Age". A good amount of research would show that the composers of the album founded Twisted Records and sell their music there. Therefore, the only real way to make sure that you're getting an authentic CD, would be to buy it from Twisted Records and not from any other obscure publisher.

A great example of this is God Lives Underwater's Up Off The Floor. The official release was very obviously sourced from MP3.

[frozenspeed]

That's a good one.

I was referring to some of the titles on the Century Media label available on Amazon.

Care to mention which ones? I only ask because I have a bunch of Century Media cds and now I want to make sure they're not in my collection

[r0k]

Hi. I've read this topic with some interest especially since i'm now trying to buy my digital music in FLAC and i want to be sure i'm not fooled by some company selling mp3 recoded into flacs.
I've ran a little test comparing the SoX spectros of 2 flacs for the same song. Once was coming from such a store, the other one was coming from ... well, you don't want to know
I got this

Can anyone give me a clue why i got so many differences. It doesn't look like the lossy paterns showed in a previous post so i guess both a re lossless, but it looks like they come from 2 different versions of the same song. I don't think this album was re-mastered.

Oh, BTW, i can't tell wich one comes from the store because i don't know if SoX will put the first file at the top or bottom, and i can't analyse more, nor less than 2 files with the spectro. Any SoX guru out there who can help me clarify this?

This post has been edited by r0k: Oct 9 2011, 19:04

[soviet123]

...From what I've read, you can use Goldwave (a piece of software) to check the frequency graphs (much like post #35 above). Since lossy never (or almost never) reaches above 20k, it should be easily discernible.
It's what they say in China, anyway, since there are so many "bad lossless" sources online there.

[r0k]

...From what I've read, you can use Goldwave (a piece of software) to check the frequency graphs (much like post #35 above). Since lossy never (or almost never) reaches above 20k, it should be easily discernible.
It's what they say in China, anyway, since there are so many "bad lossless" sources online there.

Most interesting is post #6, Rotareneg clearly explained what to look for. I just tried to use the idea on my own files but i'm still unsure how to read the graphs.
And since i was analysing two files, i thought the 2 graphs were from both files, but now i wonder whether they are for left and right channels of a single file. Which one?
SoX really needs some tutorials in addition to the man pages

[Northpack]

Can anyone give me a clue why i got so many differences. It doesn't look like the lossy paterns showed in a previous post so i guess both a re lossless, but it looks like they come from 2 different versions of the same song. I don't think this album was re-mastered.

For me they look like sourced from two different lossy encodes, the latter being normalized to a higher level.

...From what I've read, you can use Goldwave (a piece of software) to check the frequency graphs (much like post #35 above). Since lossy never (or almost never) reaches above 20k, it should be easily discernible.

That's wrong. Lowpass filtering below 20kHz is often a feature of lossy encoding but not necessarily. On the other hand, lowpass filtering is often deliberately utilized during mastering, to get rid of high frequency noise and distortion, so there are many tracks showing a low pass filter which were not subject to lossy compression.

You should really try the AuCDtect tools I told you about. They are much more advanced than any spectrogram guesswork by eye.

Also note that if a file shows signs of lossy compression it isn't necessarily the distributor to blaim. Nowadays many musicians pretty carelessly use lossy formats to record and trade samples and whole tracks (in fact they do for quite a while, think of MiniDisc recorders). Some artists, like NIN, have used MP3-like artifacts as artistics means.

This post has been edited by Northpack: Oct 10 2011, 10:30

[r0k]

You should really try the AuCDtect tools I told you about. They are much more advanced than any spectrogram guesswork by eye.

Unfortunately, this application only seems to works on physical CDs. I was trying to analyse FLAC files i bought as digital downloads.

EDIT : what makes you tell they were lossy? I tried looking for "holes" in the spectrum as on post #6 rather than just a pass filter and didn't see those. Is there something else to look for?

This post has been edited by r0k: Oct 10 2011, 10:51

[Northpack]

Unfortunately, this application only seems to works on physical CDs. I was trying to analyse FLAC files i bought as digital downloads.

That's right, that Tau Analyzer frontend really should be updated to support analyzing lossless files. However, there's the command line tool and there's the foobar component to analyse files. Or, if you want to use the frontend, you could make a virtual image from your FLACs.

EDIT : what makes you tell they were lossy? I tried looking for "holes" in the spectrum as on post #6 rather than just a pass filter and didn't see those. Is there something else to look for?

Of course it doesn't look like the spectrum from post #6 because that shows a time scale of ~4 seconds whereas your spectrum shows a time scale of ~260 seconds. You wouldn't look at a sattelite image of the country you live in and wonder why you can't see your house on it, wouldn't you?

[r0k]

That's right, that Tau Analyzer frontend really should be updated to support analyzing lossless files. However, there's the command line tool and there's the foobar component to analyse files. Or, if you want to use the frontend, you could make a virtual image from your FLACs.

Ah. I'll try to get the command line tool as soon as their web page stops giving me errors.

Of course it doesn't look like the spectrum from post #6 because that shows a time scale of ~4 seconds whereas your spectrum shows a time scale of ~260 seconds. You wouldn't look at a sattelite image of the country you live in and wonder why you can't see your house on it, wouldn't you?

Well, of course i would first take a magnifying glass
Silly me, i didn't notice the timescale on the first screenshots and wrongly guessed this was an entire track

EDIT : Northpack : Thanks a million times for pointing my sillyness.
I've re-analysed my files, selecting a 5 seconds portion, and here is the result :

File i got in torrent


Legal file from Qobuz


This one clearly has been lossy - compressed.
I'll check by re-downloading the files as wav but i feel like they are stealing us, especially since they charge an extra fee for "lossless".
Looks like i'll keep buying cheap mp3 and then downloading flacs from torrents when i can't get a physical copy of an album

This post has been edited by r0k: Oct 10 2011, 11:39

[r0k]

Not sure why i can no longer edit my previous post, maybe it's too old.
OK, before i cause any harm, i got spectrograms for track from 3 other albums i bought at Qobuz (i had bought 4 already), and none of them shows the lossy pattern visible in the previous spectro. Unfortunately i don't have other files to compare.
It might be that the files they received from the publisher were lossy to start with. I won't stop buying flac as i said in my previous post, but i will keep checking them. Hopefully this one was an isolated accident.

[dhromed]

Have you notified them?

[Northpack]

I've re-analysed my files, selecting a 5 seconds portion, and here is the result :

The one from Qobuz is from a lossy source for sure. But the first one is definetely not from the same file as one of the spectrograms in your first post. Are you sure those spectrograms are not just left and right channels from the Quobuz file?

[r0k]

I havn't contacted them yet because i first wanted to check the .wav file they offer shows the same pattern. It does.
My first spectro was actually the 2 channels of the Qobuz file. I finally understood what i had to write in the command line to get the spectro of one file. I actually forgot to add a -n parameter for the output file, wich caused quite some havoc.
The joys of command line tools

BTW, since someone asked, here is what i used

CODE

sox <"Input file name"> -n trim <x> <y> spectrogram

Where <"Input file name"> is the name of the track you want to analyse. Better put it under quotes as chances are great you have command line unfriendly characters in the name (if only a simple space).
<x> is the time where you want to start analysing.
<y> is the time you want to analyse. In my case was 5 (seconds).
If you omit trim <x> <y> you will analyse the whole track. This can be useful to spot the area you want to check more precisely.

Edit : do i need to add i strongly suggest you run SoX on copies of you tracks, in a separate folder. Cause i should have done this

This post has been edited by r0k: Oct 10 2011, 16:01

[_m²_]

You should really try the AuCDtect tools I told you about. They are much more advanced than any spectrogram guesswork by eye.

Sadly, auCDtect has MANY false positives.

Unfortunately, this application only seems to works on physical CDs. I was trying to analyse FLAC files i bought as digital downloads.

Just decode them to wav or, like suggested, use some frontend that will do it for you.

[r0k]

By "False positives" do you mean files wrongly identified as lossy or wrongly identified as lossless (or both)?

Just decode them to wav or, like suggested, use some frontend that will do it for you.

Decoding won't help me with Tau Analyser as it works only on CDs. I would have to burn the files to a disk to analyse them (or create an image and mount it), i can't just analyse files on the HD.

I found the command line tool, but it's in the "old" downloads so i guess it havn't been updated in a while. Didn't found the foobar plugin.

Well, at least i know how to use SoX now.

This post has been edited by r0k: Oct 11 2011, 17:36

[Northpack]

By "False positives" do you mean files wrongly identified as lossy or wrongly identified as lossless (or both)?

AFAIK wrongly identifies as lossy, not the other way around. I got false positives for tracks which had a heavy noise reduction filter applied to them - but it's really hard to discern such filters from lossy compression because they work on similar psychoacoustical principles.

This post has been edited by Northpack: Oct 11 2011, 17:34

[_m²_]

By false positive I mean that the file is fine, but detected as lossy.

Tau Analyser uses the same engine as auCDtect, so you can decode to wav and use auCDtect directly or automate it with fooCDtect or another frontend.

[Joseph93]

3 things:

1. hi I am new

2. I recently stumbled across a paper which details an algorithm that, with a very high success rate, guess the bit rate of an audio file just using data from the file's high-frequency spectrum. If developed further it could remove the need to visually inspect a spectrogram etc. and would be much faster.

http://www.fileden.com/files/2009/2/14/232...20Frequency.pdf

3. If anyone has Max, I threw together a patch the other day which allows you to compare the quality of two audio files in realtime (by looking at their spectrograms; they play simultaneously). I successfully used it to tell the difference between files of varying bitrate, even between 256kbps and 320kbps mp3s. you can also use vst spectrograms with it if you prefer. If anyone is interested I can upload it

This post has been edited by Joseph93: Oct 12 2011, 05:14