Skip to main content

Topic: Trying to learn to understand spectral analysis of mp3s (Read 6419 times) previous topic - next topic

0 Members and 1 Guest are viewing this topic.
  • pelmazo
  • [*][*][*][*]
Re: Trying to learn to understand spectral analysis of mp3s
Reply #50
I forgot to mention synesthesia
Mentioning won't do. I know synesthesia. I don't see what it has got to do with our topic. If you want to convince anyone, you will have to actually present a coherent argument.

Quote
And for everyone still bearing a "Your eyes cannot hear" tattoo on their foreheads I prepared some links:
None of those links appears to have anything to do with our topic. I consider it quite rude to expect us to read through long texts without even a hint as to what you think those texts are contributing to the topic.

Quote
I operate with [live by] the first statement when acquiring new music. Which I think cannot be argued with, in regard to high frequencies
MP3 drops high frequencies because its underlying psychoacoustic model judges them to be relatively unimportant for the perceived sound. If the psychoacoustic model is good, the judgment is correct, no matter what your favorite dogma may say. The psychoacoustical model that MP3 uses has been tested and refined a lot, and the hearing of very many people went into it. You may have the idea that the frequencies between 16 kHz and 20 kHz have a significant bearing on how you perceive music, but you are almost certainly mistaken.

Quote
The second one is based on what music I've already have and the new that happens to have cut-off lines below 19-19.5 kHz
You won't find many followers here if you don't show some evidence that you can discern this in practice. Merely repeating your strongly held belief is only going to get you kicked out.

Quote
Having tried new software for making 224-256 MP3's I'm having difficulty to tell the difference between them and 320's
No surprise.

Quote
Yes. First I listen to, then maybe suspect, and only then check it out. Around 90% of the time I suspect correctly; which is in regard to that old music / codecs. I have a vast collection of music accumulated throughout the years, but only recently started to nice bad and terrible quality of some of it. And spectrograms are giving me a proof of that
Proof of what exactly? Bad quality? That a codec was involved? That the codec was old?

Quote
I will be glad to participate in such evaluation of my skills and claims: that I can tell if MP3 is of lower quality than ~256 [as I have never claimed that I can hear the difference between lossless and MP3 320]
Go on then. Learn how to ABX using foobar2000.

Quote
Am I to provide some song that I know, but in lossleess format; that will be encoded by you to lower-quality-but-made-to-look-in-spectrogram-like-320 so that I cannot cheat? Should / can I provide a rock / pop song, a classical piece and maybe something electronic?
If you do the ABX testing properly it won't matter if you can discern it from the spectrogram or not.

Quote
I can also provide you with a sample of a lossless file that looks A-OK in spectro, buts sounds terrible; which is rare case of something I do not comprehend yet
Noone disputes that files can sound bad while looking OK in a spectrogram. That's one of the easiest things to comprehend, so no idea what you are having problems with.

  • bennetng
  • [*][*][*][*][*]
Re: Trying to learn to understand spectral analysis of mp3s
Reply #51
Of course we can simply use ABX, but the problem is that zWore also likes to look at the spetrograms. If he cannot differentiate two or more files by listening he can judge the quality by his eyes as he claimed. The test is aimed at his claim about judging quality by looking at spectrograms.
Yes. First I listen to, then maybe suspect, and only then check it out. Around 90% of the time I suspect correctly; which is in regard to that old music / codecs. I have a vast collection of music accumulated throughout the years, but only recently started to nice bad and terrible quality of some of it. And spectrograms are giving me a proof of that
Obviously I am not asking for a proof if you can actually hear the differences in DBTs before looking at spectrograms.

I am interested to know his guessing accuracy.
I will be glad to participate in such evaluation of my skills and claims: that I can tell if MP3 is of lower quality than ~256 [as I have never claimed that I can hear the difference between lossless and MP3 320]

Am I to provide some song that I know, but in lossleess format; that will be encoded by you to lower-quality-but-made-to-look-in-spectrogram-like-320 so that I cannot cheat? Should / can I provide a rock / pop song, a classical piece and maybe something electronic?
Not in this way. Please read my previous posts, don't attempt to change the rules to favour yourself.
  • Last Edit: 27 July, 2016, 02:12:44 AM by bennetng

  • zWore
  • [*]
Re: Trying to learn to understand spectral analysis of mp3s
Reply #52
[...]Please read my previous posts, don't attempt to change the rules to favour yourself.
I did; and followed the link

So what you are saying is, that unless a trustworthy third party will validate that I had no chance of cheating, any tests will be futile?

And so...
[...]
Learn how to ABX using foobar2000.
[...]
...what I can do is to run some tests just for myself?



And on the side note: there are some substances, that allow you to turn music into multiple sensory experience. And there are people [like myself and my close relative] that can get nice feeling shudders when listening to music while being completely sober. Also, anyone who was not exposed to the wonders of music while in lucid dream state, should not give in to narrow minded statements abut how music is perceived
I forgot to mention synesthesia


And for everyone still bearing a "Your eyes cannot hear" tattoo on their foreheads I prepared some links:

http://www.medicaldaily.com/your-brain-music-how-our-brains-process-melodies-pull-our-heartstrings-271007

http://www.cog.brown.edu/courses/cg195/pdf_files/fall07/Gaver-howdowehear.pdf

https://www.mcgill.ca/mpcl/research

None of those links appears to have anything to do with our topic. I consider it quite rude to expect us to read through long texts without even a hint as to what you think those texts are contributing to the topic.
I invoked them so that my not so clearly verbalized claim [you do not only hear with your ears] would not get stamped on right away. They surely are not in the topic of tests that we speak of now

  • bennetng
  • [*][*][*][*][*]
Re: Trying to learn to understand spectral analysis of mp3s
Reply #53
[...]Please read my previous posts, don't attempt to change the rules to favour yourself.
I did; and followed the link

So what you are saying is, that unless a trustworthy third party will validate that I had no chance of cheating, any tests will be futile?
In your posts you stated you can use spectrograms to guess the quality of lossy encodings, without referring to the original lossless file and any possible transcoding in between.

Therefore in the quiz:
[1] the lossless file,
[2] the intermediate file (if transcoded),
[3] encoder settings like command line parameters or screenshots,
will be chosen by me and encrypted without disclosing the password until the quiz is finished.

The a,b,c,d,e... wavpack files I mentioned in my previous post will be converted from mp3 256k or less as you mentioned 320kbps mp3 is transparent to you. I am not familiar with stuffs like p-value and statistics but if no other members show up and say the quiz is too easy, I can agree you can actually see and/or hear the quality of different files if you can list at least 5 files' quality from best to worst without any errors. I think it is fair to use 5 files because you can use both your eyes and ears during the quiz. My definition of "quality" is based on the bitrate of "unknown lossy format" or the final mp3 bitrate. Only one of them will change in each set, for example:

[1]
a: lossless > unknown (x kbps) > mp3 a kbps
b: lossless > unknown (y kbps) > mp3 a kbps
c: lossless > unknown (z kbps) > mp3 a kbps
...or

[2]
a: lossless > unknown (a kbps) > mp3 x kbps
b: lossless > unknown (a kbps) > mp3 y kbps
c: lossless > unknown (a kbps) > mp3 z kbps
...or

[3]
a: lossless > mp3 , x kbps
b: lossless > mp3 , y kbps
c: lossless > mp3 , z kbps
...

Notice that for any format and any bitrate, I can use any encoder and settings. That means each file from "mp3 a kbps" in [1] and each file from "unknown (a kbps)" in [2] may use different encoders and/or settings even though their bitrate and format are same.

The best thing I can do is to choose the music files which are suitable for the quiz. Take the quiz or not is up to you, but any answers you provided will be considered as final and I will not accept any excuse after I disclosed the passwords.
  • Last Edit: 07 September, 2016, 10:06:54 PM by bennetng

  • greynol
  • [*][*][*][*][*]
  • Global Moderator
Re: Trying to learn to understand spectral analysis of mp3s
Reply #54
Here are three samples for our experts to rank in order based on audible sound quality.
Is 24-bit/192kHz good enough for your lo-fi vinyl, or do you need 32/384?

  • bennetng
  • [*][*][*][*][*]
Re: Trying to learn to understand spectral analysis of mp3s
Reply #55
@zWore
My set as described in my previous post. You can request another set with different music genres after answering this set. If 32-bit float test files are unnecessary I can post 16-bit flac next time, or 24-bit if you like.

SHA1 of orch-answer.7z to ensure the file is not corrupted:
a14dc825c76294c8849b5f4c9b3975cdf5ed08bc

  • bennetng
  • [*][*][*][*][*]
Re: Trying to learn to understand spectral analysis of mp3s
Reply #56
Here are three samples for our experts to rank in order based on audible sound quality.
Just listened. It is hard, but the spectrograms are clearly different! Should I answer it? No... If I can't even get this set right (only 6 combinations?) I must be pathetic, or deaf. I am not going to accept this.
  • Last Edit: 08 September, 2016, 08:09:24 AM by bennetng

  • bennetng
  • [*][*][*][*][*]
Re: Trying to learn to understand spectral analysis of mp3s
Reply #57
Another set. Now the wv files are in 20-bit due to lack of space, but I added a -1dB gain in all files to avoid clipping. Since the answer archive also includes the original lossless file and mp3 files, it is easy to use null test with -1dB gain to verify the differences.

Due to zWore's infrequent activity in this thread, I would set a longer deadline until the end of 2016. Members who are interested can reply to this thread in 2017 and see the answer.

SHA-1 of electro-answer.7z:
25c9095e20b1b49ce20c6e931c5b134245d17e7c

  • pelmazo
  • [*][*][*][*]
Re: Trying to learn to understand spectral analysis of mp3s
Reply #58
I invoked them so that my not so clearly verbalized claim [you do not only hear with your ears] would not get stamped on right away. They surely are not in the topic of tests that we speak of now
Wenn do not only hear with our ears, indeed. The brain is also involved. We don't need to read long texts to know that. We're not as dumb as you seem to believe.

This doesn't bring us any nearer to understand what point you are trying to make.

  • greynol
  • [*][*][*][*][*]
  • Global Moderator
Re: Trying to learn to understand spectral analysis of mp3s
Reply #59
infrequent activity
That's one way of describing it.

Similar behavior can be found here and here.
  • Last Edit: 09 September, 2016, 09:36:25 PM by greynol
Is 24-bit/192kHz good enough for your lo-fi vinyl, or do you need 32/384?

  • zWore
  • [*]
Re: Trying to learn to understand spectral analysis of mp3s
Reply #60
[...]
Therefore in the quiz:
[...]
I don't think I understood half of what you wrote


So, from best to worst?


Set I: E, A, B, D, C
The so called classical music in general is where the the low quality stands out
If had to bet by money, I would most definitely bet on C being the worst


Set II: A ,C, D, B, E
Electronic music in general is where the the low quality pretty much hides out
This was hard. I don't listen to that kind of music almost at all


infrequent activity
That's one way of describing it.

Similar behavior can be found here and here.
Maybe you do not have to work and get to spend all day on any forum that you've ever made a single post on, but I have also other activities that I need to attend to

  • zWore
  • [*]
Re: Trying to learn to understand spectral analysis of mp3s
Reply #61
[...]
The so called classical music in general is
[...]
Electronic music in general is
[...]
Oh s**t, I made an assumption without giving proof of such claims, here comes the HydroPolice to beat me up for not writing something like "based on my experience of evaluating old files"

  • bennetng
  • [*][*][*][*][*]
Re: Trying to learn to understand spectral analysis of mp3s
Reply #62
So, from best to worst?

Set I: E, A, B, D, C
The so called classical music in general is where the the low quality stands out
If had to bet by money, I would most definitely bet on C being the worst

Set II: A ,C, D, B, E
Electronic music in general is where the the low quality pretty much hides out
This was hard. I don't listen to that kind of music almost at all

orch password:
Golden eyes? Wooden ears!
d: 256k
b: 224k
a: 192k
c: 168k
e: 147k

electro password:
if (Shmup != STG)
b: 256k
d: 238k
e: 209k
a: 192k
c: 169k

As you can see in the answer archive all files were directly encoded from lossless sources to mp3 without transcoding, so it should not be too hard for you, right?

The orch set should be easier as I did not use too many customized settings in the encoders, I was hoping you can at least partially right but to my surprise your answer is totally unexpected, you voted e as the best one?

Some of the electro files were encoded using previous version of LAME with deprecated switches which can generate beautifully looking spectrograms in low bitrates.
  • Last Edit: 28 October, 2016, 08:50:16 AM by bennetng

  • Dark_wizzie
  • [*][*][*]
Re: Trying to learn to understand spectral analysis of mp3s
Reply #63
This thread is getting hilarious. :)

  • bennetng
  • [*][*][*][*][*]
Re: Trying to learn to understand spectral analysis of mp3s
Reply #64
In fact even a people without any knowledge in lossy audio can detect a lot of bullshit in this thread. After I posted a 61kbps aac with full spectrum he said it sounds bad, and commented my electro test sample hides out low quality. The funny thing is that both files are mostly synthesized. Without referring to the lossless source how can someone determine the sound quality of synthesized music? What if the original file also consist of some subjectively unpleasant sound components? What if the low bitrate files hide such unpleasant sound and make them sound better to someone?

For that reason I also posted the orch set since people who familiar with orchestral instruments may be able to identify the artifacts, the test sample even included triangles, which often used as killer samples but he voted the smallest file as the best one, what can I say? According to TOS8 bitrate is simply a parameter, so it implies higher bitrate not necessarily sounds better but hey, he is the one who said mp3 is non-transparent until 256kbps, I only adjusted the test conditions according to what he said.

I think the quiz can be simplified by only post the hashes without the actual archive before revealing the answer. The hashes are just used to ensure I cannot change the answer after the challenger replied to the quiz. And to people who want to replicate this method in other forums, beware of some evil moderators may silently change the hashes in your posts and nullify your efforts!

  • greynol
  • [*][*][*][*][*]
  • Global Moderator
Re: Trying to learn to understand spectral analysis of mp3s
Reply #65
 He ignored my samples. I'm hurt. :(
Is 24-bit/192kHz good enough for your lo-fi vinyl, or do you need 32/384?

  • bennetng
  • [*][*][*][*][*]
Re: Trying to learn to understand spectral analysis of mp3s
Reply #66
Maybe he was too busy in other activities and forgot to check your samples! Just hope one of these activities is not spreading false knowledge on the internet or real world.

  • greynol
  • [*][*][*][*][*]
  • Global Moderator
Re: Trying to learn to understand spectral analysis of mp3s
Reply #67
Just hope one of these activities is not spreading continuing to spread false knowledge on the internet or real world.
Fixed that for you.
Is 24-bit/192kHz good enough for your lo-fi vinyl, or do you need 32/384?

  • Northpack
  • [*][*][*][*]
Re: Trying to learn to understand spectral analysis of mp3s
Reply #68
Hilarious topic indeed.

Whever someone suggests that he or she could determine the quality of a track by looking at the spectrogram I answer by asking a simple question: "Can you dance to spectrograms?".

Re: Trying to learn to understand spectral analysis of mp3s
Reply #69
The hilarity really kicks in for me with comments like this...
No... If I can't even get this set right (only 6 combinations?) I must be pathetic, or deaf. I am not going to accept this.
Having perfectly normal and scientifically explainable perceptual limitations makes a person human, not pathetic or deaf. Struggling to accept these limitations at first is also human, so excusable except in the face of clear evidence to the contrary. Welcome to the human race. :)