MP3 OR NOT

[dorrcoq]

Originally Posted by irishcrazy2005

Just out of curiosity, what is wrong with the tools in EAC? Also, are there any good freeware/open source spectral analysis programs out there? I don't do any actual taping or audio remastering (nor do I know anything about it!), so I don't feel the need to shell out $$$ for CEP or Audition. Thanks!

-Phil

you could do what I did - download the free 20 day trial version of CEP. When the 20 days are up, you won't be able to do much else, but all the spectral analyses still work!

[kuba]

How can you tell a difference between ATRAC and MP3 sourced material just by looking at spectral analysis?? I mean, not every MP3 file is cut right above "xy" kHz, there are various encoders and bitrates, so it can be easily cut at "xy+2" or "xy+3.756" kHz... Do you see my point?

[tannis]

What would you guys say that this one was.

[tannis]

ok , im trying my best to learn here and i have been back through all the related threads on the forum , so far ive gathered that blocky & haircuts =l lossy , therefore the analysis i posted above is clearly mp3 , just a couple of questions for the experts on here as ive recieved a job lot of discs and to me everyone of them is suspect in one form or another , now obviously i dont want to clutter up the thread with loads of attatchments so if someone could help here that would be appreciated.

1: i have a couple of discs where the frequency hits the wall at 18,000 , a little too high for mp3 i would have thought but they also have a hint of a haircut about them , whats the likely source for these.

2: if it's a minidisc recording what is the easiest way of determing that from an mp3 , i.e is there a distinct cut off point for a md recording compared to the 16,000 for mp3 , as a couple of these discs seem to have a drop off at around 14,000.

as you can probably gather by my wordings of the questions i dont know too much about this so be gentle

[ssamadhi97]

Well.. the "16kHz cutoff for mp3" thing is sort of a myth which does not even nearly hold true for all encoders and all bitrates.

It is generally "expensive" to store high frequencies, and we can't really hear much of them in "real world" / "real music" situations, so lossy formats in general tend to discard these first to save bits and use them to encode the things we actually CAN hear as good as possible. For most current audio codecs this means the lower the bitrate, the lower the cutoff will be.


Additionally, mp3 has a tendency to cut off at 16kHz most of the time even when there's a decent amount of bitrate available for encoding.

One problem of the mp3 format in general is that it is disproportionately inefficient at encoding signals above 16kHz, this is why any decent mp3 encoder avoids to encode such content as long as possible. Only if there's plenty of bitrate available or the signal is really strong above 16kHz the encoder will preserve these frequencies.


I guess the "mp3 = 16kHz cutoff" is related to the fact that for a long time the lion's share of mp3 files out there was encoded at 128kbps, a bitrate at which most mp3 encoders stop caring about content above 16kHz at all, because preserving any of it would be an insane waste of bits that are much better spent elsewhere (namely on encoding the content below 16kHz as good as possible)

[ssamadhi97]

Originally Posted by kuba

How can you tell a difference between ATRAC and MP3 sourced material just by looking at spectral analysis?? I mean, not every MP3 file is cut right above "xy" kHz, there are various encoders and bitrates, so it can be easily cut at "xy+2" or "xy+3.756" kHz... Do you see my point?

On file format and encoder "fingerprints":

There are several characteristics that can be used to tell which lossy format (sometimes even which encoder, when it comes to mp3) was used to process an audio file.


Here are the (in my opinion!) most important things you can look for:

• The cutoff frequency: an obvious tendency to cut off 16kHz usually equals mp3 (as explained in my previous post).
• the "shape" of the cutoff
• "blocking": pretty much all lossy formats require that samples are processed in blocks of a certain length. Most formats use only one or two different block lengths, so-called short blocks and long blocks. The length of these differs between formats. On a close-up screenshot of a spectrum you can estimate the block lengths and thereby identify formats.
• stray blocks due to different decisions on which block to encode with which degree of detail

I'll fill in some blanks in this post next time I feel like it. To be continued...

[Five]

Originally Posted by tannis

What would you guys say that this one was.

this is most certainly lossy, as evidenced by the blocks. you can also tell it's VBR (variable bit rate) by the jagged blocky pillars. CBR (constant bit rate) tends to cut straight across.

I'm pretty confident about spotting lossy source (thanks especially to ssamadi97's posts going all the way back to STG), the tricky thing for me these days is identifying MD since this is the one lossy source that has wide acceptance amongst traders. If the source is pre-90's and blocky, it's certainly not good for sharing (at least not here or any of the other popular traders' sites).

This screenshot looks like it could from an MD source. From what I've seen, MD resembles other lossy VBR codecs with very fat "dash-like" blocks. btw, what is this show?

ps thanks ssamadhi97 for the excellent and informative posts (as per usual).

[Five]

okay, back to the original topic, vicsbitz has sent me a 5.6 second sample of the original show in question. at this point I think this is mp3-sourced, not MD-sourced.

Frequency Analysis:

[Five]

5.6 second Spectral Analysis:

[Five]

2 second Spectral Analysis:

[Five]

if anybody wants a copy of the SHN sample, just pm me.

[Tate]

This is a good thread, very informative (we've got some real experts here )

Concerning the original post, I'd say it's MP3-sourced by looking at the zoomed-in spectral view. That area that I circled (sorry about stealing your pic Five ), that's typical of MP3.

One thing I did a while back in order to spot MP3s is I took a bunch of different MP3s I had and converted them to WAVE files, just to familiarize myself with the spectral view

[ssamadhi97]

Originally Posted by Five

you can also tell it's VBR (variable bit rate) by the jagged blocky pillars. CBR (constant bit rate) tends to cut straight across.

Where did you get that idea? I doubt that it's possible to distinguish CBR from VBR using a spectrogram



^^ CBR

Originally Posted by Five

This screenshot looks like it could from an MD source. From what I've seen, MD resembles other lossy VBR codecs with very fat "dash-like" blocks.

..and regular MD (ATRAC) is a CBR codec too, at a constant 292kbps

[Five]

I drew my conculsions from studying these screenshots that kotti put on STG ages ago. Check these out:

[Five]

...