Skip to main content

Notice

Please note that most of the software linked on this forum is likely to be safe to use. If you are unsure, feel free to ask in the relevant topics, or send a private message to an administrator or moderator. To help curb the problems of false positives, or in the event that you do find actual malware, you can contribute through the article linked here.
Topic: Undecided, please help (Read 5936 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

Undecided, please help

Hello everyone,

I am having difficulties in choosing between two different sets of files, both encoded with LAME at 320 kbps.
I'm talking about an entire discography and the pattern repeats itself, more or less as illustrated by the 4 images below.
Source is supposed to be the same (Super Audio CDs).

Supposing that one couldn't give a listen, judging strictly from a spectogram analysis, which of the following would you say is better (audio-quality-wise).

Very much interested to hear  your thoughts, thank you very much.

P.S.

The 2nd set of images (2 & 2x) display a less dramatic difference in spectral imagery.




Re: Undecided, please help

Reply #2
Message removed.
Clips were longer than 30 seconds.
korth

Re: Undecided, please help

Reply #3
If you don't know exactly how the files were encoded. You either like how they sound and accept them AS IS. Or you have to TRUST that the encoding was done as best as possible.

If accept AS IS, and TRUST are not enough. Then you'd have to obtain the album in a lossless format, and do the encoding yourself.

There's likely to be newer encoding software in the future, and the results don't have to be bit-by-bit the same, but it doesn't make it better or worse than a previous result, unless it actually sounds better or worse.

Re: Undecided, please help

Reply #4
Where are the file sets from?

Hard to judge by spectrograms.

By using the given images, I would say, the one with the stronger lowpass is the better one because it has less bit-rate wasted for less important frequencies. But this is less important at 320kbps.

But you don't know how the files are made. Probably the one file is encoded from the other one or both are made from an poor 128kbps source.

What you can do:
Take an audio editor and remix the channels to mid/side and listen to the side part. The one with more artifacts may be the worse one.

Zoom into the spectogram and look for holes in the high-frequency ranges. Many holes=worse.

Just listen to the files. If you hear no difference and both originate from the same master, then don't bother. If different masters are used, they are different by nature so keep both.

Recode them by yourself to 96kbps and listen to them to find out for what artifacts you have to listen for. Then try to spot these in the original files. If that fails, the files are just good enough.

If one of the file sets are coming from Google Play, trash or refund it. I had terrible experiences with that store, because they recode 320kbps even from very poor sources (Lame V4 with disabled lowpass and fast encoding and then ran trough MP3Gain for trackwise normalisation). I didn't trust my ears, because the display said 320kbps, until I started to analyze them and the journey ended soonly in some residual traces for the V4 thing (forgotten meta tags)... So the source where they are from is not unimportant. There are fake 320's out there. Even in official stores. :-(
- I abandoned this account since I didn't find a way to delete it -

Re: Undecided, please help

Reply #5
I didn't trust my ears, because the display said 320kbps, until I started to analyze them and the journey ended soonly in some residual traces for the V4 thing (forgotten meta tags)...

If you mean mediainfo output, e.g. "Encoding settings: -m j -V 4 -q 3 -lowpass 20.5" then you're wrong. Such info doesn't mean that the file was encoded with -V 4.

Re: Undecided, please help

Reply #6
I have yet to get acquainted to the rules around here, apologies everyone for the confusion on my part.

Here be the files,

they are both from the same master, a Remastered Edition of the whole Dead Can Dance discography, a 9 DISC Set of SACD.

The non-"x" files are encoded by someone else using LAME 3.92.
The "x" files are encoded by me via Audacity using LAME 3.99.5 from somebody else's FLAC.

This is the source of my confusion, given the fact that LAME 3.92 is much older than LAME 3.99.5.
There is no way I can replicate the non-"x" spectrogram using 3.99.5 + Audacity.
I kind of fail to understand why and am relatively worried that I'm using the wrong codec (despite being the newer one).

At a first glance, the LAME 3.92 encoded files seem to have more information.
One would say it's the less important information but I don't know what to believe.
Is LAME 3.92 the best LAME encoder to date?

That's why I'm asking for your interpretation.

Do I hear any difference between the files?
I'm biased already, I sometimes believe I can "hear more" on the 3.92 ones because there's more information there.

 :( 


P.S.

The exact same thing happens with the non-Remastered version of the original CDs.
They're all encoded with LAME 3.92 by someone else.

I've the FLACs and encode them with 3.99.5 and can see the pattern replicating.

Re: Undecided, please help

Reply #7
ABXing is an unbiased way to tell if you can hear a difference or not at all. Blind = unbiased.

A developer will usually say their latest stable release is their best release.

Lossy codecs are designed to lose information that doesn't matter so they can focus on what really matters. And high frequencies that humans can't normally hear is a good place to shave off some bytes, so that those bytes can be used where it does matter.

Re: Undecided, please help

Reply #8
The treble is quiet in these samples in range of -80 .. -95 dbFS, below threshold of audibility in presence of louder sounds, which is why it was removed by the encoder. Refer to the colors and included scale, very convenient to do on paletted spectrograms from SoX and IrfanView, which can show the color index number under cursor. Another track with loud crash cymbals would have that range preserved.

In general, if you want a "full spectrogram" in L,R use q normal (5) or default (3) and joint stereo. At 320 kbit the encoder will be very conservative with use of JS anyway. Or use another codec like wavpack lossy.

Re: Undecided, please help

Reply #9
Quote
If you mean mediainfo output, e.g. "Encoding settings: -m j -V 4 -q 3 -lowpass 20.5" then you're wrong. Such info doesn't mean that the file was encoded with -V 4.
Why?

Quote
The non-"x" files are encoded by someone else using LAME 3.92.
The "x" files are encoded by me via Audacity using LAME 3.99.5 from somebody else's FLAC.

Both sets have equal amount of information -> 320kbps. The art of lossy conversion is to decide, which information can be discarded to fit the bit-rate.

I assume that the FLAC's are true lossless and not bloated lossy files.

Trash those suspect "non-x" and don't bother with them anymore, if you don't know, how these were made.

In Audacity, you have to set the dithering setting properly before you export:
16-bit source without editing that alters the waveform -> off
Higher bit-depth of source or editing or resampled -> on ("triangle" should be enough)

Dither is a "good" quality loss (very very very low noise). It prevents distortion in some circumstances. If you just recode from 16-bit to MP3, dither is unnecessary.

Unless there are extremely quiet parts, this loss is usually inaudible, but is clearly visible in the treble area of a spectogram. That may fool you but is in fact inaudible as long the music is playing without cranking up the volume knob during silence. It is not THAT issue if you set this setting wrong. But it makes spectograms very dusty.

And after you set it properly: Make fresh MP3's from the FLAC files with proper settings and the most recent LAME version.  :))

Is your result is still worse than you expect, you should analyze the FLAC files. It might be possible, that these are made from lossy files and your old MP3 files are made from different source. But if you encode from original CD using a proper tool, the MP3's are as best as possible. If that is not enough, use different codec (Opus, Vorbis, AAC...). But don't think too much about the upper half of the spectogram that suggests "more information". That is only the treble area where human ears can't hear that good. The most unimportant frequencies are here. The really important information is in the hot bottom half, where you can't really tell by looking at the spectogram, which file has more information in that part. If the MP3 encoder decides to sacrifice the highest frequencies first, it is a good decision, even if it looks more thinned out on spectogram.  That is part of the psychoacoustic model MP3 relies on.
I think, the recent LAME has some improvements in that model so it discards more upmost treble in order to keep more mid-range.... which is better.
- I abandoned this account since I didn't find a way to delete it -


Re: Undecided, please help

Reply #11
@lvqcl:
Might be true, might be wrong... Does not really matter. It was just a trace :P .

Can you tell the bad file from the good one? https://imgur.com/a/lWvmNOn . Do both originate from lossless sources? :P
The screenshot is showing the left channel of two 320kbps files from the same song obtained from different sources.  ;D
That difference is in fact audible (especially the holes in the midrange you can see in the middle of the screenshot). It has nothing to do with Hi-Fi anymore. Looks (and sounds) like: Youtube -> Adding noise(?) -> 320kbps MP3

@TO: That screenshot can be helpful for you. It is the exact same situation: One MP3 from "unknown" source with outdated LAME version (3.90.3) (top) and the other one (bottom) is made by yourself from a lossless source. As you see, a lot of information is missing in the upper one, even in the mid-range. The missing information was already gone before conversion and LAME doesn't bring it back. That is, what a terrible-sounding MP3 looks like. Never trust an MP3 that isn't made by yourself :D.
As long as a high-quality MP3 looks like the bottom one (solid wall until at least 15kHz, a few or no floating "bricks" above) and you don't hear any artifact, be happy and enjoy the music! :)
- I abandoned this account since I didn't find a way to delete it -

Re: Undecided, please help

Reply #12
Lots of talk about holes and listening to the difference. All of it is completely meaningless when it comes to sound quality.

@Korabeu:
ABX testing is all that matters. Pay no mind to Franky666.


 

Re: Undecided, please help

Reply #13
Thank you all for your answers,

I will try to explain further the origin of my relative-confusion on this subject, very well exemplified by this particular discography that spans over hundreds of files and their different encoding result (at least in part) from the same master.
I'm pretty sure that as far "Remastered 9CD SACD Box Set" part of the discography goes, the source is the same and the FLACs I used to encode to mp3 to are true lossless files.

I understand very well that the most information in most - if not all - music is within the upper boundaries of 15khz, let's say 16khz.
There is of course variation of that boundary's importance relative to the musical genre/track played but generically speaking, information above the 16khz margin gets less important the higher the frequency goes.
I don't contest that.
I also understand that this simply is a fact,  for various reasons, one of the most important - if not the most - being the human incapacity to hear those frequencies.
I also understand that most - if not all - lossy encoders strive to achieve just that, preserve as much information possible up until that 16khz shelf, sacrificing to a lesser or greater extend (relative to encoder, its parameters, quality of source files etc.) what is above that.

As a side note, but related to the usefulness of frequencies above 16khz, I have no problem in "passing the test" described here:
https://hydrogenaud.io/index.php/topic,115880.0.html
And not to be be mistaken, I don't crank the volume up in order just to hear that 18khz tone, quite the contrary.
I have to turn down the volume quite a few decibels from my usual listening levels because I find that particular segment to be incredibly irritating.
I'll only add this, to reinforce what the original author ( magicgoose ) of that post said, Ulver as a band should really NOT be judged
by that sample.  :P


Back on track though, the "x" files seem to me to have a much harder 16khz shelf (that is replicated throughout all files) than the non-"x" files.
I know that at a first glance, the non-"x" files seem to have more information, despite being encoded with a much older LAME codec (3.92 vs 3.99.5). This is part 1/3 of the reason for my confusion.
Albeit more of the less important information (>16khz).
I want to believe that this is so in hope that that "x" files retain more information below the 16khz, the newer version of the codec being presumably smarter than the older.

But I don't see that necessarily represented in the spectrogram. This is part 2/3 of the reason for my confusion.
Someone suggested in the earlier replies that there are ways to inspect spectrograms in a more detailed/zoomed manner?
I would appreciate some help on this particular issue.
I looked for "holes" below the 16khz line in the non-"x" spectograms (LAME 3.92) hoping that this would certainly constitute the final and decisive argument in favor of the files encoded by me (the "x" files) and I couldn't really find any.
I can imagine some holes but that might not be serving my purpose.

Also, ABX testing reveals no difference to my ears. This is part 3/3 of the reason for my confusion.

So how does one decide if "spectrogram analysis" fails simultaneously to ABX testing.
Is one to believe that LAME 3.92 is better than LAME 3.99.5 though he encoded himself with the latter from 100% lossless files?

My question comes in the larger context of having to organize large to very large libraries of music, a process which I'm sure is familiar to many of us.
With the end scope of obtaining a final (be it "lossy") version of that library.
I'm hoping to achieve a far more efficient way of getting that done, other than that of spending days/weeks of deciding which mp3 versions of the same encoded FLAC is better.

Again, thank you all for your contribution.

Much appreciated.  :-[





Re: Undecided, please help

Reply #14
I want to believe that this is so in hope that that "x" files retain more information below the 16khz, the newer version of the codec being presumably smarter than the older.

But I don't see that necessarily represented in the spectrogram. This is part 2/3 of the reason for my confusion.
Someone suggested in the earlier replies that there are ways to inspect spectrograms in a more detailed/zoomed manner?
I would appreciate some help on this particular issue.
The < 16kHz region of the spectrum does appear to be more dense in the more recent encoding. My guess is that the gap around 9kHz is considered masked by the signal immediately below. You can use SoX to create zoomed spectrograms of fixed resolution that can be easily aligned and compared. Here, I selected 160 samplings per second, and isolated the stereo difference channel (it doesn't show anything unusual, because simple stereo was used) with the following command:

sox.exe "input.file" -n remix 1 2 1v0.5,2v-0.5 spectrogram -X 160 -Z -10 -y 257 -t "input.file" -o "input.file-160-lrs.png"

1 - 1x

Re: Undecided, please help

Reply #15
Sigh.

Spectral "density" is absolutely no guarantee of greater audible fidelity.  Again, to wit, Franky666 hasn't the foggiest idea what he's talking about.

To understand this requires that you understand how the encoder works, even if only on a general level.

Your time is better spent doing that instead of the fool's errand of going down this current path. There is plenty of information out there.

With regards to the pathological sample, you will learn that it is indeed possible for an encoder to decide to keep the 18k signal. Even a rudimentary understanding should be enough to know why.

Re: Undecided, please help

Reply #16
Quote
So how does one decide if "spectrogram analysis" fails simultaneously to ABX testing...

...other than that of spending days/weeks of deciding which mp3 versions of the same encoded FLAC is better.
If you can't hear a difference you can't get any "better" than that!!! 

In fact, a lower bitrate that sounds identical could be considered better compression because the file is smaller and that's the whole purpose of file compression.   (You may not care about file size, but that's why MP3 was created.)

The spectrum isn't that useful...   If you listen to a low-quality MP3 it's NOT the loss of high frequencies (or the loss of any other frequencies) that you notice.   Same thing if you hear a compression artifact in a higher-quality MP3...    Compression artifacts don't sound like low-pass filtering or EQ.

If you are using lossy compression, the sample data is going to be changed...  Every sample will be changed.     And with audio compression, the goal is not to get the data mathematically similar or to make a similar spectrum.   The goal is to get similar (or identical) sound!


Re: Undecided, please help

Reply #17
Quote
Compression artifacts don't sound like low-pass filtering or EQ.

Interesting you said that, can you develop on that please?

I use EQ in Foobar2k to compensate for Sennheiser's HD 650 "veiled" feeling.
For most music the effect caused by the EQ is truly pleasant but for some rare cases - e.g. some Dead Can Dance albums - it seems to cause voice and bass distortion.
I noticed that this is only true for some CD masters and it doesn't apply at all for some of the LP versions of the same albums.
This goes for both lossless and lossy files.

How to judge then, is it the CD's master of low quality, is the LP master the right one (though there's a significant overall volume drop), is the EQ used too aggressive etc. ?