MP3 Degradation, an experiment.

[Depone]

So, we all know that MP3's dont really stand upto raw .wav or .aiff quality. Although some argue a good 320kbits mp3 is hard to tell the diference apart.

Well, I have manflu and bored, so im hoping to show some of you without the knowledge of what mp3 encoding does to your audio.
The way in which mp3's work are very clever indeed. Instead of playing every frequency intended in that sound at once, it only plays the most 'needed' or important bit to audibly be passed off as being all those frequencies... It works on psychoacoustics.
Weird stuff... And the lower the Kbps, the more information or needed bits are lost. It mainly comes out of treble, not sure why actually, maybe someone can explain...

If anyone has anything to add to this topic I shall copy paste it here:-


Ok, so I have the example clips of various qualities and a picture of its spectrum below it. I will give a brief description of whats going on at each stage.

.wav (raw)



The audio -
http://www.depone.co.uk/mp3%20distructi ... neclip.wav

--------------------------------------------------------------------------------------------------------------------------------------

320kbps mp3

Here I can see a cutoff line appearing at approx 16khz




The audio -
http://www.depone.co.uk/mp3%20distructi ... 20clip.mp3

--------------------------------------------------------------------------------------------------------------------------------------

192kbps mp3

Here you can see that the audio is really starting to degrade, the line before has become stronger, and we are loosing more treble




The audio -
http://www.depone.co.uk/mp3%20distructi ... 92clip.mp3

--------------------------------------------------------------------------------------------------------------------------------------

128kbps mp3

Wow, what a change, just look at what the horrible mp3 encoding has done to all that lovely treble! sucks!




The audio -
http://www.depone.co.uk/mp3%20distructi ... 28clip.mp3

--------------------------------------------------------------------------------------------------------------------------------------

96kbps mp3

Well well... in this next one, the graph doesnt even reach 16khz. and its cut at 12khz... wow.




The audio -
http://www.depone.co.uk/mp3%20distructi ... 96clip.mp3

--------------------------------------------------------------------------------------------------------------------------------------

56kbps mp3

ok now we are getting into DnBTV oldskool streaming quality




The audio -
http://www.depone.co.uk/mp3%20distructi ... 56clip.mp3

--------------------------------------------------------------------------------------------------------------------------------------

32kbps mp3

and at 32..... what can i say....




The audio -
http://www.depone.co.uk/mp3%20distructi ... 32clip.mp3

--------------------------------------------------------------------------------------------------------------------------------------

Make sure that you listen to the clips provided, they demonstrate the visual loss of quality, although to me the top quality mp3 is very close to the wav...

[Grumblex]

This was a good read large up depone . hopefully all those rippers will now realise aha

[dj nation]

yea mate, nice lil read, im sure lots of ppl will find it usefull, keep on it

[Depone]

Also, just bear in mind that soundcloud streams in 128kbps quality, just image how glorious it would sound on vinyl/cd at full resolution

[3za]

Just one bone to pick, you should have kept the scales of the graphs the same, so they can be compared visually.

you should take your graph of the 192, and 128 to soundcloud, and show them how just a few kbps, can make a big difference.

[dj nation]

Also, just bear in mind that soundcloud streams in 128kbps quality, just image how glorious it would sound on vinyl/cd at full resolution

does that go for the download in soundcloud aswell?

[Depone]

Just one bone to pick, you should have kept the scales of the graphs the same, so they can be compared visually.

you should take your graph of the 192, and 128 to soundcloud, and show them how just a few kbps, can make a big difference.

Wasnt posible im afraid, the graph automatically resized to the very highest frequency thats physically there.

Also, just bear in mind that soundcloud streams in 128kbps quality, just image how glorious it would sound on vinyl/cd at full resolution

does that go for the download in soundcloud aswell?

No no... if you upload in .wav you can download in wav, it just streams in 128

[stompzi]

Instead of playing every frequency intended in that sound at once, it only plays the most 'needed' or important bit to audibly be passed off as being all those frequencies... It works on psychoacoustics.

...

It mainly comes out of treble, not sure why actually, maybe someone can explain...

You had it, right there. One of the main curiosities they found in studying auditory masking was that higher frequency sounds are harder to distinguish in the presence of lower frequency sounds (I assume down to a certain point, though? I r not audio scientist ). So, if you're gonna cut something out...

[legend4ry]

Also, just bear in mind that soundcloud streams in 128kbps quality, just image how glorious it would sound on vinyl/cd at full resolution

does that go for the download in soundcloud aswell?

Nah, soudcloud downloads are of the original file uploaded.



An interesting experiment and you really do notice the different when you listen to a 128kbs and a 320 straight after...It just goes to show with storage being so cheap you have no excuse to listen to anything less than 320 quality on your computer as you'll be losing out on so much and clearly won't be appreciating the piece of music for all its worth.

[Depone]

Instead of playing every frequency intended in that sound at once, it only plays the most 'needed' or important bit to audibly be passed off as being all those frequencies... It works on psychoacoustics.

...

It mainly comes out of treble, not sure why actually, maybe someone can explain...

You had it, right there. One of the main curiosities they found in studying auditory masking was that higher frequency sounds are harder to distinguish in the presence of lower frequency sounds (I assume down to a certain point, though? I r not audio scientist ). So, if you're gonna cut something out...

Right, that makes sense. Juxt position? or is it that lower frequencies have simpler waveforms, and somehow seen as more important?

[Depone]

Also, just bear in mind that soundcloud streams in 128kbps quality, just image how glorious it would sound on vinyl/cd at full resolution

does that go for the download in soundcloud aswell?

Nah, soudcloud downloads are of the original file uploaded.



An interesting experiment and you really do notice the different when you listen to a 128kbs and a 320 straight after...It just goes to show with storage being so cheap you have no excuse to listen to anything less than 320 quality on your computer as you'll be losing out on so much and clearly won't be appreciating the piece of music for all its worth.

Its such a shame that a lot of really good music, I can only get hold of in 128kbps. Not talking about dubstep, but some old funk and disco I only have in 128 - 192 quality... and are usually ripped from vinyl, so transcoding takes place as well... double par!

[Ongelegen]

nice post depone

[Shekul]

What would be the Kbps equivalent of all those Dnb mixtapes i had back in the day? Or would does that not make sense as tape is entirely different obviously...Anyway, i sure don't miss the days of crappy tape.

"Nah i like track 3 the best"
*Fast forwards tape for about 2 minutes, stops then plays to hear which song the tape is at*
"Fuck i went past it"
*rewinds a bit*
"...this is the end of the first, fuck it ill just play from here"

sigh

[Depone]

What would be the Kbps equivalent of all those Dnb mixtapes i had back in the day? Or would does that not make sense as tape is entirely different obviously...Anyway, i sure don't miss the days of crappy tape.

"Nah i like track 3 the best"
*Fast forwards tape for about 2 minutes, stops then plays to hear which song the tape is at*
"Fuck i went past it"
*rewinds a bit*
"...this is the end of the first, fuck it ill just play from here"

sigh

I actually used to love he sound of my prodigy cacettes. Had MAXBASS on standard.

[back2onett]

One of the best ways to understand why it's the treble that suffers most from low quality is to look at additive synthesis; most good additive synths will use as many harmonics as possible to give the most accurate approximation of a waveform. If we look at a square wave for example it doesn't take all that many harmonics to get a recogniseable square sound:



Even in this diagram we can see a fairly distinct square shape forming already after using 4 harmonics but additive synths like sytrus (If you want to look at this yourself this is a good synth to use because it can easily break down any wave into harmonics) all 256 available harmonics are used. As more harmonics are used the waveform becomes more square-like



Looks pretty square already and that's only 15 harmonics.

By adding more high frequency sine waves we can better approximate the sqaure wave (the wavey bits eventually will look straight) until the point where it looks perfectly square



49 harmonics and you've got a convincing square.

So let's say we're happy with this square and don't need any other harmonics, so that's over 200 high frequency harmonics that we can get rid of to make a more lightweight nimble audio file (MP3)

For the most part it's not noticable until you get to sounds where it is very high frequency heavy (drums especially suffer from this) or until you start really compromising on quality (less than 128kbps at least)

This post might seem a bit rushed, I just grabbed some pictures from wherever so they don't fit together. I know a bit about the Fourier series and basic DSP but I'm sure one of out resident mathematicians can explain in more depth.

[sinehere]

Very interesting, I have compared Wav and 320 many times and I still can't tell the difference, sometimes I think I can but I then realise it's placebo. It is quite easy to visually see the difference with frequency spectrum programs though as your experiment suggests. Good one Depone

[tavravlavish]

Also, just bear in mind that soundcloud streams in 128kbps quality, just image how glorious it would sound on vinyl/cd at full resolution

does that go for the download in soundcloud aswell?

Nah, soudcloud downloads are of the original file uploaded.



An interesting experiment and you really do notice the different when you listen to a 128kbs and a 320 straight after...It just goes to show with storage being so cheap you have no excuse to listen to anything less than 320 quality on your computer as you'll be losing out on so much and clearly won't be appreciating the piece of music for all its worth.

some old funk and disco I only have in 128 - 192 quality...

I WANT

hook a bruddah up

[3za]

Also, just bear in mind that soundcloud streams in 128kbps quality, just image how glorious it would sound on vinyl/cd at full resolution

Vinyl = full resolution

But seriously I would like to see how the vinyl holds up.

[staticcast]

To be honest, 320kbps mp3 shits all over your average DJ vinyl setup for "fidelity". Yes, that includes your brand new Ortofon Pro QBert Super Gold cartridges. The fact is, both formats introduce their own artifacts. Those of vinyl sound - most people think - quite nice. Those of mp3 don't, but - as I've said many times before - I'm skeptical of anyone who thinks they'd be able to *really* tell the difference between a properly-encoded (not iTunes) 320kbps mp3 and a wav on your average home studio setup. The artifacts introduced by the vinyl cutting process are orders of magnitude more noticeable than those introduced by encoding to 320kbps mp3. Again, the point is that these artifacts sound "nice", whereas mp3 artifacts don't.

Dep, your analysis glosses over quite a salient point: the psychoacoustic nature of MP3 encoding applies not only to the frequency domain but to the time domain (for example, a quiet segment immediately after a loud transient, a certain frequency when there are certain other frequencies playing loudly, a certain frequency when certain other frequencies were playing loudly half a second ago - etc etc). Each MP3 encoder has a different decision process to determine which parts of the signal can be "fudged" with fewer audible artifacts, and that's what makes some encoders better than others. It shouldn't really be a surprise that there are some nasties hidden when you go and take a fourier transform of the signal -- the whole point of lossy compression is that it sounds better to the human ear (or eye) than it looks under the unflattering eye of the microscope. How many of you are gonna be pacing around the next DMZ with a handheld spectrum analyser and a concerned look on your face?

(I'm playing devil's advocate here because I love and buy vinyl. But seriously, just use your ears.)

BTW, if you want a more immediate and obvious illustration of where MP3 encoding cuts its corners, try listening to the side signal in M/S mode (ie flip the phase of one channel and then sum them).

[staticcast]

49 harmonics and you've got a convincing square.

So let's say we're happy with this square and don't need any other harmonics, so that's over 200 high frequency harmonics that we can get rid of to make a more lightweight nimble audio file (MP3)

For the most part it's not noticable until you get to sounds where it is very high frequency heavy (drums especially suffer from this) or until you start really compromising on quality (less than 128kbps at least)

This post might seem a bit rushed, I just grabbed some pictures from wherever so they don't fit together. I know a bit about the Fourier series and basic DSP but I'm sure one of out resident mathematicians can explain in more depth.

Kinda, but not really. You can only represent harmonics up to the Nyquist frequency, so no digital representation of a square wave will ever look like a perfect square; there'll always be wiggles. (Either that or a rounded edge, depending on what you do with the frequencies just under Nyquist.) Actually, if you draw a "perfect" square wave in a wave editor and then run it through a spectrum analyser, you'll see that there's probably a significant amount of aliasing.

What you're describing is actually just low pass filtering. I'm sure you know what that sounds like. MP3 encoding works quite differently.