Jump to content
IGNORED

Checking the real quality of your music files


Recommended Posts

Some of you may be interested in a Mac OS X program called SoundScope (find it on Version Tracker as a demo download, costs 99$ if you want to buy it). This program has two interesting audio graphical analysis tools, an FFT which is able to show the spectrum of the input music (volume dB versus frequency), and a spectrogram which shows a moving display of frequency spectrum and volume versus time.

 

If you playback a track to the SoundScope and look at the FFT & Spectrum displays you can see the width of the audio signal content and it dynamics. I did this by installing the Mac essential piece of freeware called Soundflower, which allows you to redirect sound from any program output to any program input (and copes with up to 32bit/96kHz bandwidth). Then I played iTunes out to Soundflower and select Soundflower as the input to Soundscope. I set up Audio MIDI to ensure that Core Audio output (thus of iTunes) of the Mac OS X was at its best at 24bit/96kHz (restart iTunes after changing this setting).

 

And what a surprise you will get when playing different tracks. What I have noticed is

 

1. MP3s and the original iTunes AAC DRM protected tracks, have VERY limited frequency content, often not much higher then 14-15kHz (so much for 29p tracks from Amazon). and the compression of the tracks tends to compress also the volume range of the audio, thus losing any wide dynamic impact. The sound is thus “squashed” and unrealistic. The later iTunes AAC purchased tracks (no DRM and 256kbps) are a little better, but still have a very sharp cut off of their frequency response at around 18-19kHz, and tend to compress the dynamic range.

 

2. CD and 16bit/44.1kHz downloads have a better frequency response, of course. But the sample frequency of 44.1kHz means that the best theoretical frequency which can be recovered by a DAC is 22kHz, and in reality to have an accurate reconstitution of the waveform it is a lot less than that. The FFT of 16bit/44.1kHz audio shows a very, very sharp cut off of the frequency content at around 18-19kHz just where the sharp cut off filters come in on the DACs. The spectrum display also shows a very sharp cut in the high frequency content and a kind of compression of the frequency domain. This of course is why sound stage, dynamics and instrument quality cannot be reproduced by CDs - no matter how good the player and DAC (so stop worrying that you bought a cheap DAC!). The reproduction will always lack the fast transients which need a much wider bandwidth to pass them, for example the dying of a cymbal crash or the breath of woodwind or the screech (may I say that) of a violin. There was, I noticed, and tendency for some commercial tracks to have a poor signal to noise ratio and higher noise floor, which was compensated by the recording engineers by compressing the sound dynamic range...

 

3. 24bit/96kHz (I cannot display higher sample rates as the Mac audio system is limited to 96kHz). BUT wow, for a well recorded track the difference in the FFT and Spectrum is dramatic. No sharp frequency cut off of signals in the musical source spectrum, all of them are passed through, right up to 40-50kHz - well outside the limits of human hearing. Thus the reproduced music audio has in it those transients which are produced by the original instruments and the sound quality is much improved. The sound stage is cleaner, the stereo positioning better, the shear musicality of instruments improved...

 

But I have noticed quite a few artifacts in some recordings. Low level tones at high frequencies, for example 16 and 32kHz, presumably coming from the ADCs? These appear as sharp peaks on the FFT and horizontal lines across the spectrum displays. In some cases they come and go, which I can only assume is when a mix was made of different takes in a piece using possibly different recording setups. I have also seen some tracks which claim to be HD or high quality, but in fact are just 24bit/96kHz up sampled from CD tracks. Be careful of this con trick.

 

Anyway, now is the time to check the quality of your music media you have on your computer, and I am willing to bet you wish you had not, and you might even start deleting some files.

 

 

Link to comment
  • 2 weeks later...

But I always use Goldwave. It's not free, but because of the way the unlicensed version is setup you can probably use at for a year for free.

 

s., Sadly your observations are true. Be careful though with the roll off seen at 19K because IMO this is just some I-don't-know-what-trick at digitizing (hence "mastering") because this is not always so. If you recognize your DAC doing it, ok, but I recognize it as something which well might be "used" in half of the albums (at using a filterless DAC).

 

I'm currently working on analysis software myself, and at testing I already ran into the most strange things. What about cutting off amplitues at the BOTTOM of of it. Yea, you hear right. At the bottom. So, think of a file (both channels) not wanting to go under a certain value. Low-clipping, or whatever name I'm going to give it myself.

 

Anyway for others : keep in mind that *any* straight line in the spectral analysis (running in real time !) is an anomaly.

 

Peter

 

Lush^3-e      Lush^2      Blaxius^2.5      Ethernet^3     HDMI^2     XLR^2

XXHighEnd (developer)

Phasure NOS1 24/768 Async USB DAC (manufacturer)

Phasure Mach III Audio PC with Linear PSU (manufacturer)

Orelino & Orelo MKII Speakers (designer/supplier)

Link to comment

If my memory serves me right SoundScope was a software used in speech pathology/audiology in a university that I was associated with but I maybe wrong since I have not seen it in more than a decade. As such it may or may not be appropriate for the analysis you propose. If you are using any audio analysis software I would recommend that you read it's specs and limitations as it can produce its own errors and display artifacts if it is being used outside the intended function. I don't think that the vagaries of modern music creation will surprise anybody familiar with the subject as heavy compression and heavy filtering/equalization are par for the course. Good quality, downloadable acoustical recordings preferably covering the mp3, CD, HR spectrum would provide better insight as to what is possible. In the end I don't think that there is a debate about which format is more accurate but what is audible to the human ear. How about stopping some artists/producers/engineers from fooling around with all the buttons on a DAW because they can :-)

 

Link to comment

I´m doing "remasterings" of normal CD audio for years now. The sharp frequency cutoff is not new to me. Furthermore, it is needed in order to supress aliasing artifacts mirroring back into the audible signal. Sometimes you have the cutoff at 19 kHz, sometimes precisely at 22.5 kHz. It depends on the sample rate converter used during mastering or the ADC during recording.

 

Transients and room impression can be reconstructed by a fair amount when upsampling to 24/96, I´d say for 60-70% with the right algorithms. On http://www.thesoundtrackzone.com/viewtopic.php?f=35&t=323 I gave an example.

 

E-MU 0202 USB wired with Monster USB Cable --> Audioquest King Cobra --> (sometimes) Corda Arietta --> Sennheiser HD-600

Link to comment

... take a look on the Steinberg Wavelab series.

 

Even an older (second hand) version will do the job.

I have used V.4 for years doing these kind of analyzing my recordings and ripps.

Now I'm using V.6 with WavPack support, amd try to get what compression method I use to store my CD collection on harddisk ... ;)

 

The above mentioned "spikes" at around 16 and (sometimes) 32 khz are mostly from video-monitors. But sometimes there are also sideband-artefacts from AD convertors or other outboard equipment. Usually these are well down in level, so they "should" not harm, but who wants to say for sure ;)

 

i.e.: the L downloads offer some good examples or what shoul/could be done with highrez recording (the same goes to accusence or the HDTracks demos)

 

Greetings

Harald

 

Esoterc SA-60 / Foobar2000 -> Mytek Stereo 192 DSD / Audio-GD NFB 28.38 -> MEG RL922K / AKG K500 / AKG K1000  / Audioquest Nighthawk / OPPO PM-2 / Sennheiser HD800 / Sennheiser Surrounder / Sony MA900 / STAX SR-303+SRM-323II

Link to comment

It is impossible to see if a flac file was made from an mp3 even with softwares like that. I have compared both an mp3 and a FLAC and they are almost identical, very hard to categorize the differences.

 

MacBook Pro > M2Tech Evo > Stylos SYS HAD > Sovereign Director > Sovereign Power > Tidal Piano Cera (Cabling: Argento)

Link to comment

Hi wavac!

 

I would not say that this is "impossible".

It might be hard, though. Depending on the Bitrate and encoder quality (of mp3 of course), and also on the music file itself.

 

You would have to learn to set up a FFT based analyzer the right way also (so, not that it hides the differnces you're after).

 

With some acoustical recorded music - be it classical, pop or jazz - it might be easier to "see" what is happening at the higher frequencies.

 

Cheers

Harald

 

Esoterc SA-60 / Foobar2000 -> Mytek Stereo 192 DSD / Audio-GD NFB 28.38 -> MEG RL922K / AKG K500 / AKG K1000  / Audioquest Nighthawk / OPPO PM-2 / Sennheiser HD800 / Sennheiser Surrounder / Sony MA900 / STAX SR-303+SRM-323II

Link to comment

Im saying it because I have done many comparisons on many different trancks, spectral analysis, frequency response etc. and they do not show the difference between mp3s and uncompressed.

 

Actually even by listening a well encoded mp3 256 and above its impossible to recognise it in an AB blind test. I am speaking through personal experience.

 

MacBook Pro > M2Tech Evo > Stylos SYS HAD > Sovereign Director > Sovereign Power > Tidal Piano Cera (Cabling: Argento)

Link to comment

Please feel free to post 2 screenshots of the same track in mp3 and lossless indicating the differences in the spectral analysis plot.

 

MacBook Pro > M2Tech Evo > Stylos SYS HAD > Sovereign Director > Sovereign Power > Tidal Piano Cera (Cabling: Argento)

Link to comment

The point is, that these mp3 "switching" could be seen easily in realtime.

A screenshot would only face you a single moment, whi9ch could be very misleading!

 

But I've made two shots, the first is the original *.wav file, the second is a 256kbps mp3 (lame, encoded with wavelab 6).

It is not exactly on the same position, as I have made the pictures "on the fly" ...

 

WAV:

44.1khz-16bitwavxbfi.gif

 

MP3:

256lamemp3zvuq.gif

 

I think it should be easy to see, what I (and Peter) have been saying.

 

Cheers

Harald

 

Esoterc SA-60 / Foobar2000 -> Mytek Stereo 192 DSD / Audio-GD NFB 28.38 -> MEG RL922K / AKG K500 / AKG K1000  / Audioquest Nighthawk / OPPO PM-2 / Sennheiser HD800 / Sennheiser Surrounder / Sony MA900 / STAX SR-303+SRM-323II

Link to comment

Yes it is obvious in this picture that above 16khz there isnt much info...Maybe you are right, with audacity I was using I couldnt spot any difference at all!

 

MacBook Pro > M2Tech Evo > Stylos SYS HAD > Sovereign Director > Sovereign Power > Tidal Piano Cera (Cabling: Argento)

Link to comment

Hi wavac!

 

I haven't ever used Audacity, so I could not tell you about the FFT analysing properties of that software.

Maybe there are some settings that should/could be altered to get the analysis done more accurate?

 

I use the following settings for the Wavelab 6 FFT section:

Blocksize 4096

Overlap 99%

Range 0 - -150 db

Frequency 20 - 96000 Hz (depends on samplingrate)

 

I have made presets for other frequeny ranges (i.e. 10 - 100 Hz, 10 khz - 96 khz, ...) with a higher blocksize-setting, to get a more detailed look of these bands also.

 

Cheers

Harald

 

Esoterc SA-60 / Foobar2000 -> Mytek Stereo 192 DSD / Audio-GD NFB 28.38 -> MEG RL922K / AKG K500 / AKG K1000  / Audioquest Nighthawk / OPPO PM-2 / Sennheiser HD800 / Sennheiser Surrounder / Sony MA900 / STAX SR-303+SRM-323II

Link to comment

You can’t test the quality of your music by playing it though a spectrogram...that’s nonsense.

 

If you look at frequency spectrum of music and see a high end roll-off, that’s because the music rolls off at high frequencies and not the result of bad recording!

 

If you want to compare the fidelity of different types of music codecs, you must use recorded test tones and sweeps. Unlike music, a test sweep will be recorded at a constant sound level across the frequency spectrum. Any visual differences perceptible on the spectrogram would be vanishingly small.

 

 

 

 

 

Link to comment

You guys are making it quite confusing. Ok, to me :-)

 

A 256 Kbs MP3 doesn't have a roll off at 16KHz. 192 Does. It may depend on the decoder used perhaps, and the encoder will play its role too.

This by itself is not important, but what is, is when a 256 (and 320) doesn't show a roll off, well, there's nothing to be seen. Not like this.

 

I was talking about (real time !) spectral analysis, and not an FFT graph ...

This works with colors, and it is very easy to see the "weight" of frequencies. Or that peak the above FFT shows which is recognizeable from MP3 which cuts out frequencies (like 192 does) and shifts it to another one frequency (that's the peak).

 

Note that with a good real time spectogram you'll easily see the overtones of voices etc. It just takes some time to recognize it. Now grab that Goldwave so you can be sure it can show it (because I tell you, haha).

 

Best,

Peter

 

Lush^3-e      Lush^2      Blaxius^2.5      Ethernet^3     HDMI^2     XLR^2

XXHighEnd (developer)

Phasure NOS1 24/768 Async USB DAC (manufacturer)

Phasure Mach III Audio PC with Linear PSU (manufacturer)

Orelino & Orelo MKII Speakers (designer/supplier)

Link to comment

Just not to get in a wrong direction...

 

The screenshots above are taken in real time, so they are only a snapshot of what's going on with the frequency content in a file (the file used was one of acousence's demos of Mahlers 6th btw.).

 

Don't forget that the MP3 codecs are using dynamic switching of filters in high frequncy regions!

This could be seen VERY easy on a realtime FFT (like the one in Wavelab). Even at 256kbps!

 

Testtones might not show this behaviour correctly because of that "pycho-acoustic" switching thingy ...

 

So give yourself a run, take a piece of music which has the neccessary high frequency content (like a good recording of classical music), and see whats happening.

 

Cheers

Harald

 

P.S.:

I'm not claiming that this should be seen as an "absolute test of audio quality"!

But one could see, if the recording was made in higher sampling rates, or only 44,1/48khz (as i.e. some of my SACDs o_O )

 

Esoterc SA-60 / Foobar2000 -> Mytek Stereo 192 DSD / Audio-GD NFB 28.38 -> MEG RL922K / AKG K500 / AKG K1000  / Audioquest Nighthawk / OPPO PM-2 / Sennheiser HD800 / Sennheiser Surrounder / Sony MA900 / STAX SR-303+SRM-323II

Link to comment

And in addition to what Harald just said, don't forget that 50% of recordings show a roll off anyway. IOW, use more albums to get familiar with what you see and how to interpret it.

 

Lush^3-e      Lush^2      Blaxius^2.5      Ethernet^3     HDMI^2     XLR^2

XXHighEnd (developer)

Phasure NOS1 24/768 Async USB DAC (manufacturer)

Phasure Mach III Audio PC with Linear PSU (manufacturer)

Orelino & Orelo MKII Speakers (designer/supplier)

Link to comment

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now



×
×
  • Create New...