Skip to main content

Topic: True FLAC vs. Fake FLAC (Read 65054 times) previous topic - next topic

0 Members and 1 Guest are viewing this topic.
True FLAC vs. Fake FLAC
Reply #75
use and verify by spectrum. just consider date of the record

  • incifinci
  • [*][*]
True FLAC vs. Fake FLAC
Reply #76
I downloaded a fake FLAC from a blog site, with exact cut-off on spectrogram at 16 kHz. I tell them about, and they gave me a new one, with a frequency range up to 22 kHz - but the spectrogram is very strange, in my opinion, and has a big peak exactly at 16 kHz, too. I would like to know, is there a way to create fake FLAC with so exotic spectrogram, or is the new one is true FLAC? My earthes (by the test) are good (till 16 kHz), and I could not hear a quality difference between two files, using good enough quality JVC headphones.

2 FLACs, 2 Spek images + Audacity frequency analysis image are here. Thank you in advance for help.

(This like peak seems on the 60% of songs of the full album. Disney Aladdin soundtrack, hungarian, 1992.)
  • Last Edit: 15 December, 2015, 07:06:56 AM by incifinci

  • xnor
  • [*][*][*][*][*]
  • Developer
True FLAC vs. Fake FLAC
Reply #77
From a quick analysis I conclude it's just noise added on top. Maybe their source material just isn't better..
  • Last Edit: 15 December, 2015, 07:48:10 AM by xnor
"I hear it when I see it."

  • mjb2006
  • [*][*][*][*][*]
True FLAC vs. Fake FLAC
Reply #78
The first one is almost certainly transcoded from lossy, IMHO. The 13 to 16 kHz range is very selective, 16 kHz and up is just occasional noise. The spectrogram in Audition has the look of a piano roll / low-res brick-out game.

The second one has instrumental harmonics going well above 16 kHz. Everything is smooth, no blockiness. The noise at the high end looks like shaped dither, so probably it was converted down from a higher-res source. Not a transcode, in my assessment.

Here I have zoomed in on a 4.7-second section from about 10.5 kHz to 16.5 kHz and am toggling back and forth between the lossy and lossless. Levels are not matched, so ignore the difference in brightness. I am not showing the harmonics or dither noise above 16 kHz; it's just showing the blockiness. Click on the image and view it at full res if you can.

Anyway, if you can't hear a difference, the lossy encoder did a good job deciding what to throw away, and the quality levels are by definition the same. But of course you don't want to be paying for one thing (*cough*) and getting another, even if you can't hear the difference.
  • Last Edit: 15 December, 2015, 08:38:19 AM by mjb2006

  • incifinci
  • [*][*]
True FLAC vs. Fake FLAC
Reply #79
xnor, mjb2006, thank both of you for help. Especially was helpful for me the explanation of the high end noise's origin (dither).

About the peak at 16 kHz: this was not the first time, when I met this like peak. mjb2006, what do you think, is its cause the downconverting from a higher-res source, too? (As I read in wikipedia about dithering, if I understood it well.) Or is it simply because of the quantization (signal processing)?
  • Last Edit: 15 December, 2015, 01:02:34 PM by incifinci

  • greynol
  • [*][*][*][*][*]
  • Global Moderator
True FLAC vs. Fake FLAC
Reply #80
Is 24-bit/192kHz good enough for your lo-fi vinyl, or do you need 32/384?

  • incifinci
  • [*][*]
True FLAC vs. Fake FLAC
Reply #81

Thank you for a good link!

It's unfortunate, that the cause is not find, but anyway, it's good to know, that it's not because of transcoding or ripping.

  • herkow
  • [*]
Re: True FLAC vs. Fake FLAC
Reply #82
Hi guys. After reading a million posts all over the internet , I've decided to run practical tests that help me realize if we actually can determine if a song or album is lossy or not.

The answer is...  NO. (Well, if someone uses a really bad enconder, and low bitrates, of course we can, but we must test the best possible "cheater")

There is absolutely NO WAY for music. We can hardly do it in some special situations, like when we work with a single tone, but music is NOT a tone. So, to cover all the audio spectrum, we use PINK NOISE. Pink Noise is NOT music, but is pretty close, and the best for this test.

First, the Pink Noise, created in this nice audio editor:

Then, the same wav converted to Lame MP3 V1 (192-224 kbps). We can see the cutoff at the lowpass frequency.

And finally, the wav converted to Lame MP3 V0 (NO Lowpass filter) 224-256 kbps.

If we compare the original wav vs. the -V1 mp3, we may say "dude, that's lossy". But when we compare the original vs. the -V0 all we can say is "what the f"·%!".

If we take a closer look, we can see some difference in the spectral image (bigger blue points in the image), but that difference is not useful at all if we don´t own an original sample to compare. The only parameter we can use is frequency, but as you can see, when a good compression is used, there's no lowpass filtering, and the file looks like a 100% LOSSLESS CDDA.

I hope this small contribution may help.

  • herkow
  • [*]
Re: True FLAC vs. Fake FLAC
Reply #83
Another example, with music. Here is a lot easier to see the difference, but, as I said before, we must study the WORST SCENARIO, and that's Pink Noise.

This time I've used my 1994 Disque Americ (Canada) print of... Dark Side Of The Moon.
Mr. Alan Parsons did a nice mastering job...  8)

EAC, no errors, and the wav looks like this:

And now the Lame V0

Code: [Select]
This is the script if you want to use it:
#! /bin/bash
mkdir MP3 && for f in *.wav; do lame -V0 --noreplaygain "$f" ./MP3/"${f%.wav}.mp3";

As we can see, in a "cleaner environment" is easier to see the difference, even in a V0 (no lowpass) compression. Around 16KHz there's an abrupt cutoff on the low level content.

And, very interesting, the wav seems to show a cutoff around 20KHz... And we know that is a perfect rip of an original print.

So, in conclusion, an apparent cutoff (like in this wav) or a full frequency spectrum (like the pink noise mp3 V0) are ABOSLUTELY USELESS AND IRRELEVANT to determine if it's a fake or not.

In the best case, when you see a big cutoff at 15-16 KHz you can suspect it's a lossy file, but a 20 KHz cutoff proves NOTHING.

EDIT: Lossless Audio Checker said CUTOFF to my rip. So, you can't trust this software.

EDIT2: I know it may sound too fundamentalist when I say "proves nothing". I mean, this is like the "for home use, no professional"... You can "suspect" is fake or real, but no more than that.
  • Last Edit: 02 April, 2016, 05:54:08 PM by herkow

  • j7n
  • [*][*][*][*][*]
Re: True FLAC vs. Fake FLAC
Reply #84
I don't think the static noise sample tells a whole lot. Music would be more closely approximated by a mixture noises varying in level over time, sometimes sinking below where the encoder has set the threshold, other times remaining above it. Then you'd analyze how abrupt those transitions are.

A lot of CDs, in fact most in my collection, have a lowpass between 20.5 and 21.5 kHz. The rolloff is usually smoother than the one made by an MP3 encoder. To see it better, keep the time and frequency resolutions balanced so that there are not elongated pixels, and zoom in close enough. SoX defaults of 256 frequency bands and its natural colors is a good choice. An extra 10 dB of footroom can hilight artifacts better, but isn't necessary.

The Spek spectrogram isn't very clear, and has been saved with chroma subsampling, which washes out blue detail.

Check the L-R channel for blocking. With joint stereo, this part is heavily quantized.

I can't get the Lossless Audio Checker to finish processing. It has hung with 100% cpu usage for half an hour. AuCDtect gives a rather 'binary' yes/no answer, and is easily tripped by a lowpass without any encoding. A spectrogram can give more detailed picture.

• Source "Another Brick in the Wall (Part I)".

• MP3 CBR 320

Densest possible spectrum for MP3. Some parts have no obvious artifacting. I added stereo pink noise fading from -inf to -36 dB RMS. Blocking can be clearly seen up to the 18th second, where noise power is -54 dB.

The noise you demonstrated would have the power of -24 dB and is very loud, and drowns out most of the music.

• MP3 CBR 320, dither, High, Ultra Shaping

More realistic usage than the pink noise. Blocking can be seen more easily.

• Vorbis, Q8, 247 kbit/s.

The threshold rises smoothly without a jump at 16 kHz. "Piano roll" style blocking can be seen nevertheless, up to the 15th second (-60 dB noise).

  • herkow
  • [*]
Re: True FLAC vs. Fake FLAC
Reply #85
The Pink Noise is only Pink Noise. There's no music.

  • j7n
  • [*][*][*][*][*]
Re: True FLAC vs. Fake FLAC
Reply #86
LAC.exe worked, whose download link I missed; only the GUI version didn't work.
It finished very quickly, but only reported "cutoff" for both cases.

AuCDtect found the original 99% CDDA, while the transcode was 100% CDDA, same result with normal/medium dither.

Another authentic track, which sounds great, with the following smoothly rolled off spectrum, was given a score of "cutoff" / 54% CDDA. (Paul Mauriat - "My House and the River" (1970 / 2012))

Both tools appear to be of no use. Even if they got the scores approximately right, the percentage would still not tell us anything about the distribution of the problematic areas. (Such as if some lossy coded or filtered samples had been used.)

If I doubt the lineage of an album, I set my computer to generate maximum resolution spectrograms using SoX / Frontah, walk away, and later quickly flick through them.