Detecting fake 320Kbps MP3 Options

[Blutarsky]

How to detect if a track has been transcoded from lower bit-rates to higher? Can a single spectrum analysis show fake 192 to 320 transcoding?
See this

The above shows a spectral analysis from a 320kbps track; a linear cut occurs over 20Khz. Can this be sufficient to proof that the track hasn't been faked from lower bitrates?

As a question: will lower bitrates tracks show cuts at lower Khz?

Your thoughts

[Blutarsky]

A more representative sample:

A track encoded @ 128Kbps



Now the same track encoded @ 320Kbps



You can clearly notice the high cuts in the first image, however the spectrum below seems identical.... could the spectral difference between 16/20khz be due to "induced" hiss when transoding a fake? Or what?

[dhromed]

What is that crazy reduced band of the first image, around 12KHz? Looks like an accidental EQ adjustment.

[pdq]

The spectrum in the first post shows a 15.75 kHz signal from the horizontal sweep of a TV. I could hear that when I was young, but not any more.

[db1989]

How to detect if a track has been transcoded from lower bit-rates to higher?

Rip it yourself or purchase it from a service that isn’t silly enough to transcode.

[Destroid]

Also, if the quality sucks then is good chance. Of course quality could suffer from bad encoder and/or settings, which is still the same thing in the end-- low fidelity. The chances are if you can't tell 15 times out of 16 that it isn't bad quality then is likely your quest is already over

[Blutarsky]

Also, if the quality sucks then is good chance. Of course quality could suffer from bad encoder and/or settings, which is still the same thing in the end-- low fidelity. The chances are if you can't tell 15 times out of 16 that it isn't bad quality then is likely your quest is already over

Yes it may be just a waste of time, considering I can't distinguish over 15Khz!!!! What do you think?

[fuflo]

or purchase it from a service that isn't silly enough to transcode.

devil's advocate here,


how do you find out if a service is "silly enough to transcode" if do you not test the mp3s they are selling?

This post has been edited by fuflo: Feb 22 2012, 10:54

[Porcus]

In a few discussions here, it has been reported that even well-established record labels sometimes sell CD made from lossies. Who knows how many times they might have been mp3'ed ...

This post has been edited by Porcus: Feb 22 2012, 11:11

[KMD]

Is plot three straight to 320 or a re-encoded 128 . I think it would be usefull to compare three plots , 128 encoded 320 encoded and 128 re-encoded to 320. It may be that the non mp3 encoded original did not have much over 16Khz.

This post has been edited by KMD: Feb 22 2012, 15:22

[pdq]

Yes it is noise , 70dB down.

What was this in answer to?

Edit: Never mind, I see that you have edited this out of your post.

This post has been edited by pdq: Feb 22 2012, 15:17

[Blutarsky]

Yes it is noise , 70dB down.

You mean the spikes colored in blue? If so, it is an "inflated" copy of the lower bitrate track?

[icstm]

an un-thought-through question:

what would be the main impact of such a recording:

- dynamic range
- high freqency response
- less well defined signal degradation?

[KMD]

Yes it is noise , 70dB down.

You mean the spikes colored in blue? If so, it is an "inflated" copy of the lower bitrate track?

Yes

but on second reading of your original post I hve second thoughts and ask is the third plot straight to 320 or 128 to 320 as that makes a difference.

[pdq]

an un-thought-through question:

what would be the main impact of such a recording:

- dynamic range
- high freqency response
- less well defined signal degradation?

Do you mean reencoding a low bitrate to a high bitrate?

Lossy codecs are capable of a very wide dynamic range, far greater than that of even 32 bit integer pcm data, so dynamic range should not be an issue.

Frequency response is a possibility, especially if the default settings were used. However, as pointed out earlier, one can specify the low-pass filter frequency when encoding so that is not a sure thing.

I don't know what you mean by "less well defined signal degradation", but for sure there will be more artifacts. Unfortunately, without access to the original, it is impossible to say for sure what is an encoding artifact and what is not.

[Blutarsky]

but on second reading of your original post I hve second thoughts and ask is the third plot straight to 320 or 128 to 320 as that makes a difference.

I don't know, both are supposed to be encoded at their respective bitrates. That's why I was asking

[KMD]

Maybe the codec from 128 to 320 adds noise above 16khz as an artefact, and maybe the un encoded original did not have much above 16Khz.

"Can a single spectrum analysis show fake 192 to 320 transcoding?"
If you really suspect a fake, then such a “thing” (cringe) could contain deliberately added fake noise above 16Khz and deliberately added fake products from a spectral enhancer above 16 Khz . So... no.
Actually some codecs use spectral enhancers in the playback. So its not trivial.

This post has been edited by KMD: Feb 22 2012, 17:17

[slks]

There's no accurate way to distinguish a "real" 320 kb/s MP3 (encoded from a lossless source) from one encoded from another lossy source (such as another MP3).

There was an application (I forget the name) that would analyze a .wav file to see if it had been through MP3 compression at any point, however since what we're discussing here are all MP3s of various bit rates, the test would be useless. Even this test was not always accurate - the result was a percentage confidence that the audio was either MPEG or uncompressed.

So when you're trying to distinguish, not just compressed vs. uncompressed, but "Has this audio been, at some point, compressed at a lower bitrate than it currently is -" there's just no way to tell. While it's easy to spot a lowpass filter on a spectrogram, Different encoders, settings, and versions will apply their lowpass filters differently.

Not to mention that a lowpass filter could have been applied when mastering the CD, the recording might not have a wide frequency response to begin with and might only appear to be lowpassed, or the CD could have actually been mastered from an MP3 (it happens)...

Furthermore, you can't draw a direct relationship between bit rate and perceptual sound quality, because of variances in encoders. I would take a 160 kb/s MP3 encoded with a newish version of LAME over a 320 kb/s BladeEnc circa 2001 encode any day.

There's just no way to tell.

This post has been edited by slks: Feb 23 2012, 06:20

[icstm]

Furthermore, you can't draw a direct relationship between bit rate and perceptual sound quality, because of variances in encoders. I would take a 160 kb/s MP3 encoded with a newish version of LAME over a 320 kb/s BladeEnc circa 2001 encode any day.

I had no idea things had progressed some much with encoding, even though the mp3 standard has not moved (ifaik). Interestingly, ~2001 I thought mp3Pro was all the rage...

[pdq]

It's not that lame has vastly improved since 2001 (although it has improved some), but rather that BladeEnc was always a lousy encoder.

mp3Pro was a dead end, and for some very good reasons.

[db1989]

or purchase it from a service that isn't silly enough to transcode.

devil's advocate here,

how do you find out if a service is "silly enough to transcode" if do you not test the mp3s they are selling?

You got me! It’s a fair point, although I imagine most questions like this are not from people (I don’t direct this specifically at the OP) who have charged themselves with the task of monitoring the digital-audio industry. In light of the aforementioned discoveries of lossy-sourced tracks on CDs and so on, perhaps someone should!