Can spectrogram lie about audio quality?


[I know this is not the Audacity forum, but before I go there, or look for another piece of software, I would like to hear opinions from people who maybe know something about spectrum analyzing but who happen not to use the Audacity]

I've been looking recently on some files with the usage of spectrogram in Audacity. And some strange things I've noticed

For example here is music from the same movie, both in FLAC format

File AUDIO [] was downloaded in FLAC, encoded by me into WAV [for some simple volume editing, nothing fancy] and then encoded back to FLAC. And it sound great

File MOVIE [] was extracted from a video by me into WAV [for some simple volume editing] and then encoded to FLAC. Unfortunately I do not remember what kind of file it was thus I do not know the quality of audio in that video. But what is important is that it sounds bad

[I know it is not exactly the same piece of music, but those two are very similar pieces with the highly noticeable difference in quality that I've simply came to my attention when listening to my vast music colletion]

And here is the mystery: both files show frequencies above 20 kHz; but only the AUDIO sound like it. The MOVIE files looks a little different in the spectrogram [less packed,] but having audio over the 20 kHz line it sounds more like around 15 kHz. How can that be possible?

Another example is the GAME file []

I do not remember the origins of it, but I have it described as being a ripped audio from a Nintendo 64 videogame [released in 1997]. But I know for sure that this rip was not done by me, so almost for sure it was downloaded it a FLAC format and the encoded to WAV and back to FLAC [only by mistake I would encode WAV made from MP3 into FLAC]

And the question here is: how is it possible for this file to have frequencies around 24 kHz; and at the same time to sound like something with 10 kHz [which would be expected as it do comes from an old game from a low tech platform]?

Spectrogram is suppose to show fake lossless audio files [for example MP3s encoded to FLAC]. But this screenshot bares the proof of doing the other way around- showing low quality sounds as lossless:

Detecting the real quality
[X] Incorrect Tools behavior

Perhaps you have a look at a wikipedia article about spectrum analysis:

I think that such an analysis only makes sense if you can compare an expected output with an actual output. As long as you do not know what the ideal spectrum for a certain piece of audio should look like, you cannot deduct anything from it.

Or a statement could be: yes this sounds accurately like dwarfs with a cold in a tin can, nothing to improve.

  1. Dumb question, but: did you notice that your screenshots are only the first half minute?

  2. I do not know much about spectrum analysis, but I generated files with SoX, and they look a bit more like they should. Have a look:

  3. What I know about spectrograms though, is that yes-they-can-lie. Spectrogram can show information that is inaudible, and disguise information that is crucial to quality.

Have a look at this version: . Much more treble, wouldn't you think?
Then listen to it:
Does it sound like it?

Now, your upload is 1280 pixels wide. Meaning a resolution of about 300 per second.
That means each pixel "averages 100 samples". Quite crude. And you don't even see every pixel.


The spectral view of the GAME.flac test file looks like a maximal 18 kHz sound file with a visible noisy transition band around 17 kHz, so it could also have been an original 16 kHz bandwith audio file.
Because of the upsampling to 48 kHz there has been added additional noise up to 24 kHz with level of about -60 dB and smaller (for most disco people a never hearable area).

The spectral view of both test files, AUDIO.flac and MOVIE.flac, look rather noisy in the audio area of about -60 dB and smaller, this may be also the effect of upsampling to 48 kHz.
Also the spectral view looks like as the audio has been modified by some audio software, maybe in order to enhance the psycho-acoustic feeling of wide and bombastic rooms and sounds.
It looks like there is a noise veil over all frequencies in the audio area above 8 kHz.

Because the test files have no audio parts of quiet sounds, I would have no hesitation to turn them into middle quality or maybe high quality MP3 file.



It seems that you can; at least in most cases

I did some test and also took a look at some examples: low and high quality MP3s, low and high quality FLACs, MP3s turned into FLACs, music from old games, speech etc.

I used not only Audacity but also Spek and Spectro [and they showed some differences on the same files]

So now I use Spectro, because it draws a cut off line and is has the most convenient form for me. Spectro seems to be almost always correct. And in times I'm not sure I use Spek and then Audacity. [I just wish, there would be one program that would combine best features / looks off all those three]

It took me a couple of hours and some well though material for tests [taken out from my memory], but now when I look at the spectrogram I can see what is real lossless and what is "pumped up". I even understand to some extent how those aforementioned mysteries could have happened [but I lack the proper words to describe it as all of my audio knowledge is somehow extended but at the same time quite unprofessional]

I wanted to show similar graphical features shared by AUDIO and MOVIE

As they would also look for me just yesterday. But today, after those various tests, I can clearly see in those particular SoX examples, that MOVIE in reality is something about 15 kHz and GAME is very suspicious and goes no higher that 20 kHz

The only problem now is that I have to stop, look and think. There is no automation to this process if it is to be done with accurate results; even if any of those programs could load / process many files at once and spit the results in numbers putted to some tag field- I would still need to verify that with eyes and sometimes with ears

I maybe be mistaken but it sounds just like my AUDIO, with no or little quality loss. But for me it looks suspicious, because it is to much "packed" and "pumped up"; it just looks too good. And I was guessing: you resampled my file to something like 96-192 kHz, maybe even did some bid-depth conversion too, and then encoded it to MP3 320, and then to FLAC. But I did that and what came out of this process does not look anything like that

So tell me please: how did you achieved what you have shown with this AUDIO8 spectrogram sample?

I'm curious because in my tests I have seen similar fake high quality [for example original music from old game had 11 kHz, but it "remastered" version released some time later was showing 16 kHz, but in reality had the same low quality sound]


Bit-depth conversion, yes, but that's all: SoX'ed it down to 8 bits and then up again to 16. Options -r 44100 -b8 and then -r 44100 -b16.

What happened to the spectrogram then? I actually do not know. Maybe the process just induced noise everywhere in the eighth bit, and the noise is midway up the volume scale. But they sound much more equal than than the spectrograms look, and that is my point, really.


I did the same and similar test. [And as previously, I've turned on contrast on my screen to see details]

First I converted my AUDIO sample [a supposedly genuine FLAC] into WAV and then bid-depth converted it [from 16] down to 8 and the up to 16 and again up to 24

Spek showed no visual difference between new 16 and 24 [8 is not supported]
Spectro showed minimal to very little difference between new 16 and 24 [8 is not supported]
Audacity showed minimal to very little difference between 8 and new 16 and 24
[and all showed huge difference between new 16 at the original 16]

Then I converted my AUDIO sample [a supposedly genuine FLAC] into WAV and then bid-depth converted it [from 16] up to 24

Spek showed no visual difference between original 16 and 24
Spectro showed some difference between original 16 and 24
Audacity showed some difference between original 16 and 24
[and of course none showed difference between original FLAC and a WAV made from it]

So apparently going first down to 8 and then back to 16 or higher to 24 paints the spectrum with some noise, clearly visible when compared to original file; the noise that is entirely not present when going [bid-depth converting] immediately from original [16] to 24. And that noise stays when going 16-8-24-16; but aside from that noise, that first 16 looks exactly like the last 16

And all of those files sounded to me the same [but I didn't use the double blind methodology for this; or however it is called, when someone plays you the samples when not knowing what samples is of what quality]. If there is difference in audible quality [as there should be between 16 and 8 bits], it is very very little; at least for me

What I take from this is: what really counts is the number of kHz and how the look. If 20 kHz [MP3 320] / 22 kHz [FLAC] look like on steroids [like those fake 24 bits] it doesn't matter- the quality is maintained [at least in the most audible area]; it just looks weird / artificial and thus misleading. What is important in cases like those is that you cannot fake some low quality file to look like that [all colorfully pumped up through the entire width]

But if 20 / 22 kHz files look empty but / or at the same time looks like they are cut with a knife and / or have dense spots / vertical lines- the quality is low or simply bad. A tell sign are also empty horizontal areas going through the entire width [or at least some parts of a file visible as distinctive blocks]; which can be unfortunately masked pretty good by noise. And so having some 100 dB values at 20-22 kHz do not automatically make the file good / lossless, as they can be misleading [a simple way to produce such a fake reading is to take a low quality file and pump up volume in some quiet part of it]

In theory all you need is your eyes and some experience. But unfortunately one must consider this: if during the mixing of a file / record were used some low quality samples, then this will show up even when ripped directly from a CD, misleading the viewer. [An example of this is track "Running With The Wolves" by The Prodigy]


I thought I had this figured out

But I just do not know what's happening in this file:

It is suppose to be FLAC and looks like it- but aside from that incomplete horizontal line

Spectro specifically tells that this is the cut-of line [the point where the faking begins]. But it should go all the way. And also what is above it looks quite natural. But I do hear some things in it that make me wonder, if there were compressed indeed. And there are other examples like that on that FLAC album that this files comes from

The only explanation for this that I can think of is that parts of this file were made [mixed] from different recordings / tracks. But that doesn't make much sense- why would original score, written by one composer and [as to my knowledge] conducted and sometimes recorder by him, be of different quality? Something like that is to be expected if you create music using samples [like The Prodigy does]- but not in a film music compositions


Hmm, yes, this was my first thought too.
The small noise band around 15.75 kHz looks like an audio recording which has been stored onto a video tape, or maybe, there has been involved some older analog-digital effect unit, which has introduced a part of its high frequency sampling noise. I don't know how to describe this phenomenon with the right words.
The good news is that this peep tone or high frequency noise can not be so clearly perceived by the human ear. From today's point of view, this is a not so good recording.



Hallo there...

Maybe you can take a look at this little program, and look if the problem still exist than.


Spek is an acoustic spectrum analyser written in C and C++. It uses FFmpeg
libraries for audio decoding and wxWidgets for the GUI.

Spek is available on *BSD, GNU/Linux, Windows and Mac OS X.

Find out more about Spek on its website:

Spek 0.8.2 - Released 2013-02-24 over all frequencies in the audio area above 8 kHz.


As I shown it in post 8, I'm already using Spek

After going to Audacity's forum and establishing some facts, my modus operandi will consist in most cases of analyze in Spek and Spectro, followed by Audacity and my ears. Because their spectrograms can lie / misinterpret, and they work and show result in different ways; and in way should this ever be an automated process. [I've also tried Sonic Visualiser and I will try to incorporate it into my workflow and also looked at GoldWave and SoX]

The topic is broad and complex and I have found only basic info on how to read a spectrogram


Here is some more advanced info about reading spectrograms:


Thanks to user ryerman [[X] Incorrect Tools behavior] here is some more info:,111736.0.html7


Bit late the party but this is an app much like mp3val whereby it just simply tells you if a FLAC is not a genuine one. But in some cases this is could be cased by the producer using lossy samples, low pass filters on instruments & end users encoding to FLAC or any number if things.

Lossy checker app link


I've downloaded an album from beatport lately called 'Sam Paganini - Satellite' found here.

Amazing btw.

I checked the 320kbps tracks in Spectro & Spek as that's what I only purchase and was disappointed with the graphs so I went and bought the track Sam Paganini - Rave in WAV. And as you can see the files are virtually identical. So from his standpoint that's how he produces his tracks. And he's a master one at that IMO. 7 million views on youtube for a 2 year old Techno track is no mean feat'.


WAV- Sam Paganini - Rave (Spectogram from Spectro)

320kbps- Sam Paganini - Rave (Spectogram from Spectro)

320kbps- Sam Paganini - Rave (Spectogram from Spek)

To sum up it up from me, unless you get your hands on the WAV and inspect it and encode it yourself then there's no sure way to know exactly what you get.

I thought I did until I came across this album.


Yes, being a professional and to some extent successful musician / producer implies, that you know what you are doing

But consider this:
a] He may have done it A-OK, but the company / distributor downgraded the quality; for whatever reason
b] He made a simple mistake with this album and there was to late to go back to full CD quality
c] Few yeas ago I red an article which reported the outrage of audiophiles over the practice of music company that release vinyl editions, which apparently have compressed files recorded onto them

And yes- if the samples used were of lesser quality, then the track can show up as being compressed [but it wasn't- the samples were]

So going to the full extent of those issues, even analyzing an original CD can be not enough- you would have to check the master tapes

Lately I've been checking older albums and audio tracks of movies. And there are case like of a big budget Hollywood film with a top composer on board. And you can get on the album with the score in FLACs every track with frequencies going up to 22 kHz on spectrogram, but 1 or 2 tracks stopping [although without an abrupt cut of of line] at 21. At the same goes for reissue albums of music from the 80s for example [at least for rock genre from my European country]- some tracks do not go all the way up

And here is a curious case: score from "It Follows" from 2014 []

In the movie the music have all the frequencies. On the composer websites this music is free to listen to, but in lower quality. And on the Internet you can find FLAC files which look like those lesser ones from the website, but with addition of literally painted higher frequencies. Which I doubt is the case [that some pirate made fake FLACs, from those free-to-listen files available for free]


Have you tried that tool I was linking in the last post?


Yeah the artist is the person who sends it to the label which is owned by Adam Beyer (Drumcode Records) in this case. Artists on the label are Nicole Moudaber, Joseph Capriati, Pig & Dan and I've tested their tracks in my collect and they are all peaking at the 20-20.5KHz range for 320's.

Except Adam Beyer releases Going Down [DC166] and both Prisma EP [DC89] & Satellite [DCCD10] by Sam Paganini.

So that must be artist choice or admin error.

I've emailed beatport for that track rave and they gave me a full refund so that was nice of them. :slight_smile:


Probably yes, do not really remember. And this last answer of yours @stevehero somehow must have slipped my attention; I am sorry for that

But for months now I was absolutely content with Spek. But now I am resurrecting this thread to report a peculiar finding:

Spek can show a frequency cut off line - and so can Audacity. But when Audacity does it- then there is really nothing above that line. But with Spek it is different. Some compressed files [made I do not know how] if are made extremely louder [something like 500-1000%] can show in Spek a little of something than looks like a noise above that supposed cut-off point [that something that you can see in Audacity without making them louder]

I tested this finding by re-compressing some of those files [and compressing some new ones] but in a non VBR way [thus with intention of creating a cut off line]. And so even if I made them 2000% louder in editor, they would not show that noise in Audacity and Spek [and Spetro]

So my conclusion is this: if you wanna be sure about cut off frequencies, then use Audacity. But if you want ease of use, then choose Spek

[And I can only guess that the difference, with or without such little of extremely quiet "noise", is not noticeable at all to a human hear - that is a speculation, but firmly based on how this "noise" looks in comparison to what is below it or in comparison to for example how remastered music can look like which music originally was published without full spectrum of frequencies. That noise looks like almost nothing]


And on the side note: I finally cracked a long lasting mystery of my life

The question to which I could not find an answer was: why some movies in English I understand without a sweat while [rarely] some I have a hard time to watch [hear]? I am not an English native speaker and here is what I mean: It was not a case of unknown vocabulary, it was not a case of actors speaking with accents, it was not a case of speed with which they talked. As after every sentence I could think to myself "well, that was easy - so what gives?". Well, it is simply the case of frequencies! It struck my one evening to load a movie to Spek, that I was watching and was having a hard time to understand. And there it was: a cut of line at 12 kHz

Right now most of the time I am able to correctly tell if the audio in a movie is compressed to ~15 or to ~12 kHz

And the biggest proof to my findings what in form of a remastered on Blu-ray "Gone With The Wind"- a movie from 1939. It showed a whooping 8 kHz. So it was no wonder that the music in it sounded horribly and that when a black house servant spoke in a way a stereotypical southern black servants speaks in Hollywood movies, I had to turn on the subtitles for the most part


Fakin' the funk. Just bought it. It's the best one I've used. Shows frequencies that spek doesn't.

Comparing the two, Speks resolution sucks at displaying the higher frequencies.


Just to show you.

Compared to Fakin' The Funks redition. Look at the resolution of the higher frequencies and the cutoff. It's what's called a lossy master.

Mulitple breaklines [Enters] not allowed?