IPB

Welcome Guest ( Log In | Register )

 
Reply to this topicStart new topic
How to determine the intrinsic audio quality ?, Ideas for WAV-quality measurement tool
forart.eu
post Oct 25 2011, 10:21
Post #1





Group: Members
Posts: 74
Joined: 10-December 09
From: italy
Member No.: 75798



Since Pleasurize Music Foundation released well-known TT DR o.m. seems that many peoples assumes that DR means audio quality...

To my experience, even if DR is of course important, it's not the unique parameter to instrumentally determine the "intrinsic audio quality" of a WAV file; the frequency response, for example, is a parameter to take in serious consideration, IMHO.

BTW I'm not an expert, so I wanna start this discussion to understand what parameters do you consider essential in determining the quality of a WAV file.

Thanks in advice for all contributions !
Go to the top of the page
+Quote Post
Chesteta
post Oct 25 2011, 13:49
Post #2





Group: Members
Posts: 19
Joined: 24-December 10
Member No.: 86759



I assume you are talking about some sort of program to check 'how compressed' a song is (I thought I saw another discussion regarding figuring out if a song was a 'true flac' file or a flac'd mp3); one idea may be to have a 'smart' algorithm look at the decay of high frequencies - I am not sure if this would work for all codecs (or bitrates), but in foobar2000, if you look at a 'moderate quality' mp3 on the spectrogram plugin (turn dsp off as the spectrogram shows the post processed audio) compared to an uncompressed file, the highs will be very short 'dashes' as opposed to a dash with a decay; the mp3 codec cuts off the decay. it may be possible to do an analysis of various codecs and their 'tendencies' to then give a probability of whether a song was recompressed or not - this correlation number could also be used to give an intrinsic 'sound quality' rating (obviously you would use more than just the decay of the high frequencies... some songs may not be mixed with lots of highs..)

perhaps with a database or something of what sound quality ratings the original songs received, one could know for certain whether a song was compressed (presumably there would be SOME change in the song that was compressed that could be seen)... if the program couldnt detect a difference and was looking at all of the indicators of compression that would probably indicate a very good quality compressed file, which in the end wouldnt be that bad id think?

just some thoughts smile.gif

afterthought: you could probably use a neural network of some kind which learned the difference between a compressed song and an original by taking ~1000 original songs and copies of those songs in various compression formats and then figuring out what differences there are between the waveforms... I do not know how to program (so take this with a grain of salt), but I would not THINK this would be that difficult a task? then use the resulting 'knowledge' learned by the neural network to figure out whether a non-learned song was compressed or not.

This post has been edited by Chesteta: Oct 25 2011, 13:53
Go to the top of the page
+Quote Post
db1989
post Oct 25 2011, 16:24
Post #3





Group: Super Moderator
Posts: 5275
Joined: 23-June 06
Member No.: 32180



QUOTE (Chesteta @ Oct 25 2011, 13:49) *
I assume you are talking about some sort of program to check 'how compressed' a song is
I was unsure whether it was this or some universal measure of perceptual quality, even of never-compressed audio.

QUOTE
I thought I saw another discussion regarding figuring out if a song was a 'true flac' file or a flac'd mp3
The most recent one of many: True FLAC vs. Fake FLAC. There one can find discussion of several applications that attempt to predict the likelihood that a given uncompressed (or a decompressed) file was sourced from a lossy encoding, and their various limitations. Some techniques similar to your subsequent suggestions are present in such programs, and I imagine there have been past discussions on potential refinements.

Overall, many things that could be discussed here have probably been gone over before; I say that not to discourage new ideas, but rather so that you might benefit from old ones.
Go to the top of the page
+Quote Post
DVDdoug
post Oct 25 2011, 18:22
Post #4





Group: Members
Posts: 2534
Joined: 24-August 07
From: Silicon Valley
Member No.: 46454



This gets complicated... You can measure certain aspects of the reproduction chain, but the quality of the music is largely subjective. If the dynamic range is compressed, that's an "artistic" decision. The band may have a very un-dynamic playing-style and the producers may use lots of compression. Maybe the musicians are lousy and they play the wrong notes, or maybe they have lousy instruments, or lousy microphones... A computer program can't really tell you if a song "sounds good".

Ethan Winer's Audiophoolery article identifies 4 characteristics of audio quality: Noise, Frequency Response, Distortion, and Time-based Errors. You measure can the ability of a WAV file (or reproduction sytem) to minimize these. But if you have an actual WAV file, probably the only thing you can measure is noise. Dynamic compression would probably fall under "distortion", but since it's intentional it's not really an error or defect.

You might get an subjective impression of the other characteristics, such as "too much bass", or "not enough highs", but you can't easily put a number on these things because every recording is different... different instruments, different musicians, different music, etc.

P.S.
If you are a hi-fi enthusiast (AKA audiophile), you are interested in two things - First you want an accurate playback system that can play the recording as accurately as possible (low noise, flat frequency response, etc.). i.e. You want the recording to sound exactly as it did live, or exactly like it sounded in the studio. Second, you are going to look for great recordings of great performances to play on that system.

This post has been edited by DVDdoug: Oct 25 2011, 18:40
Go to the top of the page
+Quote Post
forart.eu
post Oct 27 2011, 08:46
Post #5





Group: Members
Posts: 74
Joined: 10-December 09
From: italy
Member No.: 75798



Trying to explain better what i'm going to do: it would be great to have a tool that can strumentally measure the "audiophileness" of a WAV. (i'm not interested in measure the reproduction quality)

So i wanna understand which (strumentally measurable) parameters to take in consideration for.

At the moment i'm pretty sure these parameters are important:

- dynamic range
- frequency response

...maybe sound pressure ?

This post has been edited by forart.eu: Oct 27 2011, 08:50
Go to the top of the page
+Quote Post
2Bdecided
post Oct 27 2011, 10:00
Post #6


ReplayGain developer


Group: Developer
Posts: 5058
Joined: 5-November 01
From: Yorkshire, UK
Member No.: 409



You have two such tools available - one on either side of your head wink.gif

Blind analysis is tricky (in this case, blind = without knowing what the source should have sounded like) - and with any recording of a live event, "should have sounded like" it quite subjective any way, even if you heard the live event itself.

Cheers,
David.
Go to the top of the page
+Quote Post
forart.eu
post Oct 27 2011, 12:43
Post #7





Group: Members
Posts: 74
Joined: 10-December 09
From: italy
Member No.: 75798



QUOTE (2Bdecided @ Oct 27 2011, 11:00) *
You have two such tools available - one on either side of your head wink.gif

Right, but these "tools" don't give a numeric measurement.

I wanna try to obtain something more accurate than TT DR Meter (but with the same approach) to evaluate the "audiophile-appeal" of a WAV.

I can't believe that's impossible.
Go to the top of the page
+Quote Post
dhromed
post Oct 27 2011, 13:35
Post #8





Group: Members
Posts: 1286
Joined: 16-February 08
From: NL
Member No.: 51347



> I can't believe that's impossible.

Why?
Go to the top of the page
+Quote Post
bug80
post Oct 27 2011, 14:30
Post #9





Group: Members
Posts: 398
Joined: 23-January 05
From: The Netherlands
Member No.: 19254



QUOTE (forart.eu @ Oct 27 2011, 09:46) *
Trying to explain better what i'm going to do: it would be great to have a tool that can strumentally measure the "audiophileness" of a WAV. (i'm not interested in measure the reproduction quality)

So i wanna understand which (strumentally measurable) parameters to take in consideration for.

At the moment i'm pretty sure these parameters are important:

- dynamic range
- frequency response

...maybe sound pressure ?

The problem is, you cannot determine the 'frequency response' of a WAVE file. As the name suggest, the frequency response is always in 'response' to a reference (for example an impulse in the case of an impulse response). You don't have that reference here. You can measure the (long-term) SPECTRUM, though. But that does not say much about quality. For example, music could have a very non-flat spectrum (if that is what you are talking about, I don't know) and still sound good.

Sound pressure is also non-measurable from the wave file, because the sound pressure at the ears depend on a number of things: input file, amplfier, speakers, distance to the speakers, etc.


Go to the top of the page
+Quote Post
DVDdoug
post Oct 27 2011, 22:26
Post #10





Group: Members
Posts: 2534
Joined: 24-August 07
From: Silicon Valley
Member No.: 46454



QUOTE
Right, but these "tools" don't give a numeric measurement.

I wanna try to obtain something more accurate than TT DR Meter (but with the same approach) to evaluate the "audiophile-appeal" of a WAV.

I can't believe that's impossible.

Can a computer tell you if the Beatles are better than the Rolling Stones?

Can a computer program tell you how beautiful a woman is? I can rate her on a number-scale of 1-10, but maybe I'd give her an '8' and you'd give her a '9'. Maybe I prefer burnettes and you prefer blonds???

Can a computer tell you that a Picasso is better than a painting by a 10-year old?

This post has been edited by DVDdoug: Oct 27 2011, 22:27
Go to the top of the page
+Quote Post
forart.eu
post Oct 28 2011, 09:26
Post #11





Group: Members
Posts: 74
Joined: 10-December 09
From: italy
Member No.: 75798



Well, then, if this is true than why many peoples thinks that better DR means better quality ?

A computer can't tell if a Picasso looks good or bad, but can count colors used exactly.
So if you need to test a monitor it *shoud* be better a full-color-range image than a B/N one. Am I wrong ?

A WAV frequency-range measure could be an interesting parameter to evaluate.

This post has been edited by forart.eu: Oct 28 2011, 09:28
Go to the top of the page
+Quote Post
dhromed
post Oct 28 2011, 09:51
Post #12





Group: Members
Posts: 1286
Joined: 16-February 08
From: NL
Member No.: 51347



QUOTE (forart.eu @ Oct 28 2011, 10:26) *
Well, then, if this is true than why many peoples thinks that better DR means better quality ?


Because it sounds subjectively more pleasing for many. Some people prefer loudness. Some people require the objective effects that DR compression provides. DR can be measured, to a degree. Quality cannot.

QUOTE (forart.eu @ Oct 28 2011, 10:26) *
A computer can't tell if a Picasso looks good or bad, but can count colors used exactly.


And what would you do with that information?

Say it has 5 colors. Does that mean it is a better quality picture? A worse one? Not at all. There exist literally hunderds of thousands of monochrome pencil sketches and 1-bit inks, or even 1-bit digitized line art, that are better quality than some random fool's 24-bit photoshops.

QUOTE (forart.eu @ Oct 28 2011, 10:26) *
So if you need to test a monitor it *shoud* be better [to use] a full-color-range image than a B/N one. Am I wrong ?


(I added the [to use] because I assumed that's what you meant.)

An output device has known technical capabilities. It it advisable, obviously, to use a test signal with content that matches or surpasses the output device.

But that doesn't say anything about the quality.


You can measure many aspects of a signal, for a variety of uses, but it is not possible to claim a priori that A is better than B because A has a higher sample rate, or more high-frequency content, or bigger DR, etc. You need to listen to A and B.

Edit
As an excellent example: I often watch the art on theartofanimation.tumblr.com. There's a small thumbnail, and a link to the high-resolution version. But guess what? About half the time, the high-res version is upscaled from a smaller version. It has all the pixels in the world, but it's the same quality image as the smaller one from which it was resampled. There is no way I could have discovered this without looking at the image. I suppose software could be written that detects some tell-tale artifacts of upscaling and heavy jpeg compression, but such software is pretty unreliable and produces many false positives.

This post has been edited by dhromed: Oct 28 2011, 09:57
Go to the top of the page
+Quote Post
Arnold B. Kruege...
post Oct 28 2011, 13:59
Post #13





Group: Members
Posts: 3647
Joined: 29-October 08
From: USA, 48236
Member No.: 61311



QUOTE (DVDdoug @ Oct 25 2011, 13:22) *
This gets complicated... You can measure certain aspects of the reproduction chain, but the quality of the music is largely subjective. If the dynamic range is compressed, that's an "artistic" decision. The band may have a very un-dynamic playing-style and the producers may use lots of compression. Maybe the musicians are lousy and they play the wrong notes, or maybe they have lousy instruments, or lousy microphones... A computer program can't really tell you if a song "sounds good".


You can't get around the fact that if you want to test something, you need some reference. In the case of recording and production, the logical reference is the original music file before you started processing it. Lacking a good reference, you are automatically very limited in terms of what you can do.

Go to the top of the page
+Quote Post
forart.eu
post Oct 28 2011, 14:27
Post #14





Group: Members
Posts: 74
Joined: 10-December 09
From: italy
Member No.: 75798



OK, these links may explain better what i mean:

http://www.colour-science.com/quality%20te...%20overview.htm

http://en.wikipedia.org/wiki/Test_card

http://www.equasys.de/videotestcollection.html

As you can see video "testers" does use certain resolutions, color ranges, etc.

It could be interesting to have a tool to measure "the compliance to use certain WAVs as a tester file" for audio, IMHO.

Is it more clear now ?

This post has been edited by forart.eu: Oct 28 2011, 14:37
Go to the top of the page
+Quote Post
bug80
post Oct 28 2011, 15:18
Post #15





Group: Members
Posts: 398
Joined: 23-January 05
From: The Netherlands
Member No.: 19254



The examples you are mentioning are used for testing reproduction equipment (printers, televisions, monitors) and not the content itself (like videos). So it is still not really clear what you mean, at least to me.

Do you mean that a WAV file is only good when it uses maximum dynamic range and frequency range? I can easily make a file for you with a very high dynamic range and bandwidth, but sounds VERY bad at the same time smile.gif

This post has been edited by db1989: Oct 28 2011, 15:51
Reason for edit: no need to full-quote the previous post
Go to the top of the page
+Quote Post
Zarggg
post Oct 29 2011, 22:30
Post #16





Group: Members
Posts: 545
Joined: 18-January 04
From: bethlehem.pa.us
Member No.: 11318



Hearing is not a quantitative sense by any means. When the human brain translates pressure exerted on our eardrums to "sound," it performs a number of tricks that influence what we hear. That's why we can pick out a single conversation in a noisy room and why "2-channel signed 16-bit Linear PCM sampled at 44,100 Hz" was deemed sufficient to digitize sound.

You can't evaluate the absolute quality of sound with numbers. Yes, it will give you data about certain comparable attributes of the sound, but those attributes should not be confused with any sort of overall qualitative measure. When you know the dynamic range of two samples, you can say "this sample has more dynamic range than that sample," but you cannot say "this sound is better than that sound."

This post has been edited by Zarggg: Oct 29 2011, 22:34
Go to the top of the page
+Quote Post

Reply to this topicStart new topic
1 User(s) are reading this topic (1 Guests and 0 Anonymous Users)
0 Members:

 



RSS Lo-Fi Version Time is now: 24th July 2014 - 08:19