Mp3 or Not???

[jcrab66]

Ok, with the recent talk of different analyzers I thought I would throw some shots up here and see what some people think.. Here are some shots using Eac, CEP and Analfreq. Is it Mp3 or not???

[jcrab66]

here are the last two shots....

[Five]

mp3

but the results sure look different from program to program!

can you show us a cep spectral analysis, a full song and also a shot zoomed on on 1.5-2 seconds?

does the cep fa stay like that when you move the cursor around the cep display?

could you please post another shot of the EAC fa with the fft size set to maximum?

thanks. this is shaping up to be an enlightening thread.

[jcrab66]

Ok, here are the 3 requested shots, def looks like mp3 with the CEP Spec but the freq was sort of questionable

[paddington]

The steep 16kHz rolloff is a dead giveaway, in MOST cases. It's possible that the microphone sucked or that type of thing.

If you get 16k roll-off, you can be pretty sure it's MP3, BUT one more thing you can check beyond that is this:

When an UGLY roll-off filter is applied, like what the MP3 compression algo does, you get audible "aliasing" at the high edge of the low-pass band. LISTEN to it, and see if your high frequencies sound "watery" instead if crisp.

If you hear that mess and show a 16k roll off, you probably have MP3. If the taper used a crappy mic with bad freq response, you could still have the 16kHz roll off (as a coincidence), but you wouldn't hear the aliasing (watery high end) because no shitty filter was applied - the mic just didn't get it at all.

If I had to guess without listening, I'd say MP3.





As an aside - 16kHz is PLENTY high enough to enjoy a recording, it's the dirty steep roll-off filter on MP3s that ruin the audio. Yes, you are missing 16-20kHz, but the only things really there are harmonics and some REALLY high stuff you usually don't miss. When the you encode in MP3, the aliasing creates jagged stepped edges around 16kHz and those resonate harmonics down much lower, effectively ruining the audio well BELOW 16kHz, where it's very audible (and irritating).

[jcrab66]

thanks, i am aware of all that stuff, i primarily wanted to show how different analysis tools can give different result opinions just because of the software alone....

[U2Lynne]

Thanks James. I'm still learning about identifying mp3s, so what you wrote helps me out and makes lots of sense.

[paddington]

thanks, i am aware of all that stuff, i primarily wanted to show how different analysis tools can give different result opinions just because of the software alone....

oh, in that case, some of your shots don't help much because the scale needs to be adjusted. If it seems like some of them are showing different things, it's the scale displayed. Some seem to not roll as badly until you see that the audio modulation level is like -160dB or something :lol: . You generally won't hear much below -72dB or so.

Five has a thread posted stating that CEP 2.0 or Adobe Audition 1.1 do the best job of analyzing for this kind of thing and I agree. Just get the scale adjusted appropriately.

[Five]

It would be cool to see if we could get the same results from several fft programs by tinkering with settings.

this show it seems we need the sa to really see what's going on. the fa looks kinda flat in the highs but the drop isn't visible as with AnalFreq. EAC's sa you can see the drop at 16kHz at the highest fft setting, even tho it looks a little different.

the cutoff just above 16kHz in a straight line, this is the classic case of lossy source, since this is how 128kHz CBR mp3 displays. Also, the little legolike shapes visible when you zoom in with CEP indicates lossy.

I can't seem to figure out how to adjust the scale of the fa for cep... when I pull on the left bar it just moves up and down, and there's nothing in the advanced options I can see.

Oh, while I was doing this I discovered something.

I opened CEP, loaded in a short lossy-sourced .wav I had on my desktop and the fa looked like this:

[Five]

then, when I clicked on the display, the fa suddenly changed to look like this:

[Five]

I selected all, and the fa looked like this:

[Five]

So I guess the morale of the story is when you do fa with cep you need to click your cursor anywhere on the waveform display to get the fa window to come to life.

hey jcrab66, can you try this out with your file and see if you get a different wav display like I did?

here's the sa for same file as the previous three fa shots I posted:

[diggrd]

fa in CEP is always gonna show either what's selected or just where the cursor is. It also gives the matching musical 'denomination' which I find interesting. I'm still running 1.2, I wish I has upgraded to 2.0. Now if I could find the disc it would save me what $130 bucks instead of $230 :disbelief

[62v8]

With respect to the discussion about 15KHz and up ...
Doesn't FM and TV audio cut off at 15KHz as well?

Curious as to how do you tell the difference or does that become subjective assesment?

[Five]

diggrd, what I am suggesting is that CEP 2.0+ has a little bug where you have to click on the waveform to get the correct fa to display. Then you can select all or move the cursor and the originally displayed fa never comes back (the one with the flatline above 14KhZ).

62v8, you are correct about this. FM cuts off around the same place, but not quite so abruptly. The slope on the fa will be less steep, and in spectral view the tops of the spikes will be more uneven. Furthermore, when you zoom in you shouldn't be able to find any "blocks". 32kHz DAT cuts off in much the same way and can be checked in much the same way. TV-sourced stuff usually has the famous red stripe visible roundabout 16kHz and will cut off unevenly like FM somwhere just below that. The fa will show a spike where the red stripe is too. There will normally be at least a little bit of activity (99% noise) above 16kHz, with the possible exception of 32kHz DAT digitally transferred. These ones it's extra-important to zoom in and check for those giveaway lossy lego blocks visible in the sa.

FM, TV and 32kHz DAT all have mostly no activity above 16kHz, but they are superior to lossy sources because they are free from audible lossy aritifacts like slurring of hihats and those little noises that are like scifi movie sound fx.

[RainDawg]

Five, I believe the issue with your spectral plot is that it only shows the activity where your cursor is currently placed. The first one, it may have been at the beginning of the file where nothing was happening yet. When you clicked in the middle of the display, presumably to a spot there there is more "music", it suddenly shows you a more interesting plot.

One of the weaknesses with the freq analysis in CEP is that it will only display for one sample at a time, so even if you select a large portion, it will only display the information for the sample where the cursor is actually located. AnalFreq will run a progressive analysis, and display either the current position, or an average of what it has scanned so far.

This is why lossy stuff is so much easier to spot with AnalFreq than with CEP. Since the high end content on an mp3 file is just random noise, taking the average of it over an long period of time is going to result in a perfectly flat line, whereas each instantaneous plot (as in the pics above) is going to have some activity present. It will also accentuate that steep dropoff as it takes the average, making FM vs. MP3 much easier to definitively state. FM and TV have, as you said, a nice smooth rolloff when you take the average over the whole file.

I'm just not a fan of CEPs freq analysis tool, as it doesn't seem to really be useful for spotting sources as much. The Spectral Analysis tool is great though.

[jcrab66]

Five, I got the same results as you as far as moving the cursor and getting a different wav display...

[Five]

jcrab66: that's great! :thumbsup

RainDawg: I think you may be right, it could be the first sample. We'll get to the bottom of it somehow. I think AnalFreq is great, too, I just don't like waiting so I usually only use it when I need to.

[RainDawg]

Well, that's the thing with AnalFreq, you do have to wait because it is going to scan the whole file. There is simply no way to do it on CEP...it's only going to display a single sample analysis at a time.

But with AnalFreq (I just love saying the name) you really don't need to scan the entire file. Usually about 10-20 seconds of the audio is all you need for the high end noise to average itself to zero and the rolloff becomes defined enough to tell whether it's lossy or FM.

[ffooky]

I'm surprised no one has mentioned Audacity. You get a maximum 23 second stretch to analyse and the rolloff is always very clearly defined.

[Five]

yes, let's talk more about Audacity...

I'm going to start playing around with the SA and seeing how well it works for our purposes.

I can't seem to find FA in there, can Audacity do this?

fooky, can you please expand on this bit about the 23 second stretch? I don't quite understand.

[ffooky]

Select a length of the waveform (up to 23.8 seconds in length..hence the stretch thing) and go to the View menu. Select Plot Spectrum and Bob's your uncle. I must say I find it the easiest FA to read, though the SA is merely sufficient. I haven't really tried any of the other progs apart from EAC but I find it picks noise reduction from MPEG encoding pretty well...the roll off with NR usually seems to have a stair step effect even though the SA looks like boxed standard lossy.

[Five]

ahh yes, I've got it now!

looks great! I'm going to use it for sure next time I'm doing an analysis and see how well it works.

now if I could just remember how to cut on sector boundaries using audacity :confused:

[Five]

well, what do you know.

the time is right now.

I just visited an announce thread on EZT. This is most certainly lossy, looks like VBR.

CEP SA, zoomed in and full track view:

[Five]

Here's SA using Audacity, zoomed in and fulltrack. I can't tell anything from these shots.

[Five]

Here's SA using EAC, zoomed in and fulltrack. You can see the telltale blocky lego spikes pretty well, especially on the closeup.

[Five]

Here's FA

1) using AnalFreq with default settings on the left channel. You can sort of see stairsteps in the highs.
2) using CEP set to FFT size 1024 to match AnalFreq's settings.
3) using CEP set to FFT size 512 (default)

[Five]

And finally the FA using Audacity. Audacity's FA seems to show the suspicious stairsteps the most clearly.

1) using FFT size 1024 to match AnalFreq's default
2) using FFT size 512 (audacity default)

[Five]

and here's the EAC FA, set to FFT sizes 512, 1024, and 65536 (max). Not much useful information in these shots.

[Five]

so these screenshots are all the same .wav file analyzed to death with all the common progs used. Let me know if there's any others that work well.

In this case, CEP seems to be best for Spectral Analysis, and Audacity is best for Frequency Analysis (thanks to ffooky for the good advice).

[ffooky]

Here's SA using Audacity, zoomed in and fulltrack. I can't tell anything from these shots.

Audacity's SA isn't great but it ain't that bad...you've only got the x axis (or is it the y?) going up to 7kHz :-)

[ffooky]

I'm not 100% sure this is MPEG, though it's clearly not the recording in its untouched state...what's the source given ?

[Five]

It was this:
Blind Melon - Good Foot Workshop Demos (http://www.easytree.org/torrents-details.php?id=30478)

it looks like this Hendrix mp3-sourced show that came up a couple months ago. I think it's high bitrate VBR mp3 but that's just my best guess... whatever it is it is not tradeable. I got this thinking it might be another source but it is just a bad copy of the one I already have.

they banned it

I woke up and adjusted the scale of Audtion's SA and now it works really well! thanks. Here's the new shots:

[ffooky]

Righto. If I had to give a verdict I'd say crapola noise reduction but that's as bad as MPEG IMO anyway.

BTW, for cutting on sector boundaries with Audacity:

[Five]

Righto. If I had to give a verdict I'd say crapola noise reduction but that's as bad as MPEG IMO anyway.
I have never seen any nr plug that creates those square shapes at the top of the spectrum. Bad nr can sometimes resemble the straight cutoff of CBR but the lego city stairstep look in these shots has to be lossy. It looks almost like MD compression (altho I don't think that's it either). Maybe AAC? It looks a little like this, too:

[Five]

BTW, for cutting on sector boundaries with Audacity:
THANKS VERY MUCH!!

I've saved this to my audacity folder so I don't lose it.

Okay, so you set selection format to cdda sectors as in your screenshot, then what? Is there a way to set cue stops on the sector boundaries and split a show into however many tracks or is it only possible to split by selecting in this mode, cutting then pasting into a new window and saving each individual track?

[ffooky]

I have never seen any nr plug that creates those square shapes at the top of the spectrum. Bad nr can sometimes resemble the straight cutoff of CBR but the lego city stairstep look in these shots has to be lossy. It looks almost like MD compression (altho I don't think that's it either). Maybe AAC? It looks a little like this, too:

There's definitely something rotten going on but it's the stair step pattern of the high end roll off on the Analfreq (the name killed my kids BTW) and Audacity FAs that intrigues me. I've seen this phenomenon on several shows that *claim* to be from silvers (a recent Hendrix boot on Usenet comes to mind) and I have never been able to replicate it with any of the lossy encoders at my disposal. My experience of AAC is that you tend to get a typically steep and almost complete drop at the top end of the spectral analysis, around 16kHz and then a little island of frequencies after that, whose range and depth is determined by the bitrate at which the AAC was encoded.

[ffooky]

THANKS VERY MUCH!!

I've saved this to my audacity folder so I don't lose it.

Okay, so you set selection format to cdda sectors as in your screenshot, then what? Is there a way to set cue stops on the sector boundaries and split a show into however many tracks or is it only possible to split by selecting in this mode, cutting then pasting into a new window and saving each individual track?

There may very well be a more elegant method but I park my cursor at an XX+0000 position, one track at a time and go Edit->Select...->Start to Cursor and then File->Export Selection as WAV, Delete and on to the next one. If the last track has any digital silence then I usually cut that at a boundary and discard the last few null samples or alternatively put the cursor at the very end, select Generate->Silence and then just trim that off at the first available boundary.

[RainDawg]

Hey...I saw that ffooky :nono: .

You forget that as soon as you post, it sends out the message via email to anyone subscribed to the thread. Deleting it doesn't help ;).

[Five]

There's definitely something rotten going on but it's the stair step pattern of the high end roll off on the Analfreq (the name killed my kids BTW) and Audacity FAs that intrigues me. I've seen this phenomenon on several shows that *claim* to be from silvers (a recent Hendrix boot on Usenet comes to mind) and I have never been able to replicate it with any of the lossy encoders at my disposal. My experience of AAC is that you tend to get a typically steep and almost complete drop at the top end of the spectral analysis, around 16kHz and then a little island of frequencies after that, whose range and depth is determined by the bitrate at which the AAC was encoded.
well it's really the SA that's most alarming. lossy encoding is the only thing that causes the zoomed SA to look like it's built from squares. Just because it's from Silvers doesn't mean it's lossless... there was a Metallica '86 show on EZT directly from silvers which was mp3-sourced a while back.

and thanks for the Audacity tips. it would be cool if they would add cue stops which could be on the boundaries or not and a splitting function someday.

[ffooky]

I've just come back in to the room and my 11 year old is claiming he just sent some smilies but he looks as guilty as hell. If whatever he did was worse than that please can you PM me the details.

[Five]

no sweat it really was just smilies

[ffooky]

no sweat it really was just smilies

OK then. He says he's sorry and so am I.

On a happier note, I've just had a fiddle with Audacity...First go Edit->Move Cursor...->to Track Start and hit Project->Add Label At Selection. Then select yer boundary points one by one adding a label for each one. Once you've got labels added for them all go to File->Export Multiple... Choose your export format and location, select "Split files based on:Labels", probably safest to name by numbering consecutively and hit Export :)

Yeah, dirty silvers are certainly out there. Can you post some examples of known noise reduction treated files, the more amateurishly applied the better.

[Five]

thanks so much for this tip ffooky I will be trying this out soon!

[Five]

Yeah, dirty silvers are certainly out there. Can you post some examples of known noise reduction treated files, the more amateurishly applied the better.
...and I'll see what I can do about this as well.

[RainDawg]

I've just come back in to the room and my 11 year old is claiming he just sent some smilies but he looks as guilty as hell. If whatever he did was worse than that please can you PM me the details.
Oh, it was just smiles....no problem. I was just trying to make a joke out of it there.

[ssamadhi97]

I just visited an announce thread on EZT. This is most certainly lossy, looks like VBR.

CEP SA, zoomed in and full track view:
very typical minidisc spectrum. so it's actually 292kbps cbr.

now stop spreading these vbr myths already :p

[Five]

you're right... all along I'm thinking this looks like it was digitally extracted from MD which was sourced from the famous SHN set seeded at STG. It boggles my mind that somebody would ruin this set in such a careful way that I thought it must be vbr mp3/ogg. I'll hold off the vbr namecalling until I catch up on my homework, regardless, this is lossy as f**k.