The HD music fft atlas reference thread

[Kippyy]

For those of us not savvy enough to interpret these Audacity graphs(yours truly):

1) Any links towards interpreting these would be most useful, ie what does the x and y axis mean, what is compressed vs good dynamic range,etc

2) A short statement after each graph would be useful

 

I take "burned again" to be an unflattering comment re the Moody Blues from HD Tracks.

Thanks

 

[wgscott]

1. x-axis in both plots is time. The y-axis in the first plot is volume (amplitude), and in the second is frequency (with false-color indicating amplitude as well -- stronger is more red, weaker is blue).

 

In the first Moody Blues plot, it appears that the dynamic range has been exceeded. In the second plot, there appears to be no signal above 22 kHz, consistent with a redbook CD. A 96 kHz sampled recording should in general have some signal up to half the sampling frequency, i.e., 48 kHz.

 

I don't think there is an easy way to see compression just from looking at the plot, but it is quite possible that this has been up-sampled from 16-bit, 44.1 kHz.

 

I actually was hoping that this would be primarily an atlas, a reference, rather than commentary. The idea is to see what you are getting before you buy, rather than be told something, so that if I am mis-interpreting the data, you don't have to take my word for it.

 

The best link is probably the Audacity documentation.

 

 

[Kevin_O]

There is a tutorial on graphs and such over at WBF forum by the guy that does the hi-rez transfers for HDtracks.

 

 

http://www.whatsbestforum.com/showthread.php?2749-Listening-with-your-eyes-How-to-read-graphs.

 

[ron spencer]

Did u ask hdtracks about the moody blues release? I bought it as well...pity your graphs show this

 

[wgscott]

I only bought it last night. Besides, it isn't like they have ever bothered to make it right for me before.

 

[new_media]

Is it possible that some of these HD downloads are being transferred from analog tape and that there may not be any signal above 22 kHz, or should you see noise up to the 48 kHz threshold regardless?

 

[wgscott]

The spectrum doesn't look artificially cut off (unlike the dynamic range), so that could certainly be the case. But, then, why make it 96kHz?

 

[ted_b]

with the Moody results (and anything else associated with Audacity). Bruce did the transfers from SACD (as always). He is claiming, for the nth time, that Audacity is flawed, buggy and shows inaccurate results. It's hard to argue: his FFT's are published and show much more energy.

 

 

 

 

 

 

 

[wgscott]

Well, I'd rather there be a problem with Audacity then have wasted $17. This will be a good control to see if there is such a problem, or if it is just naive user error (I am very much a naive user).

 

Edit: I can get the frequency spectrum to look similar if I crank up the gain 30 db:

 

 

So I think it is more a question of what defaults to use rather than Audacity being buggy. However, the amplitude plot worries me more.

 

[wgscott]

Here is the same track as what he has displayed, using gain settings (+30db) that seem to put them about on the same footing. I think the Audacity plot actually makes it look a wee bit better, probably because the other program doesn't display anything under 10kHz (which is probably why it doesn't normalize in the same way):

 

 

vs.

 

 

[wgscott]

This is John Coltrane, Lush Life, from HD Tracks, purchased a couple of weeks before they yanked it without explanation. It is supposedly 96kHz sampled, which means it should go to 48 kHz. You can see it abruptly truncates at about 22 kHz. Unfortunately (for me at least), no amount of renormalization will make it look any better.

 

 

Having said that, it does sound good.

 

[Miska]

With Audacity (or any other spectrogram tool) it is good to use 144 dB range for 24-bit material and 120 dB range for DSD-material, to somewhat reflect the dynamic range of the material. IOW, good starting point is a setting where the quantization noise is just visible.

 

If some part of the spectrum is flat-empty uniform color, then settings are too low and some low level parts may be cut off.

 

 

[new_media]

wgscott,

 

There is a free spectrum analyzer for Mac called Sonic Visualizer.

 

http://www.sonicvisualiser.org/download.html

 

Maybe you could you have a look at some of these same files with this program and see how the results compare.

 

[wgscott]

They all seem to give the same results, but again I think Audacity makes it look better:

 

 

[goldsdad]

"So I think it is more a question of what defaults to use rather than Audacity being buggy. However, the amplitude plot worries me more."

 

I absolutely agree.

 

 

Audacity reportedly had a problem with spectrograms in an earlier version, not the current recommended version 1.3.12-beta. Perhaps Bruce at Puget should be made aware of this improvement in Audacity.

Bruce's spectrogram can show down to -150dBFS. Audacity can show down to -144dBFS; that's the range of a 24-bit file.

 

Try the settings in the attached image.

'Window size: 4096' gives most precise frequency calculation. (It does lack temporal precision compared to smaller 'window size', but since your graphing several minutes over the relatively narrow width of a computer monitor, that lack is immaterial in this case.)

'Range (dB): 144' shows entire dynamic range of 24-bits.

 

 

 

 

 

[wgscott]

kinda' funny... there is a resident "expert" over on the CA forum spreading some bad info and thought I'd nip it in the bud. They're using Audacity, that has a known bug and they're listening with their eyes again. Below are the graphs for the new Moody Blues album on HDtracks.

 

There is an exceptionally simple solution to the problem of having meddlesome ignorami using software that doesn't display something the way you want to see it:

 

If the vendors would be so kind as to post the Fourier spectra of their high resolution tracks, we wouldn't be left to do it on our own. Then they can present everything in the most favorable light, and we wouldn't have to rely upon all these horrid, ignorant and meddlesome know-nothing self-appointed "experts".

 

By the way, my hearing cuts out at 16kHz, so forgive me for "listening with my eyes." I should just trust the authorities to tell me what I should hear.

 

[wgscott]

gain = 0

dynamic range = 144 dB

 

 

 

This gives the best-looking result as well.

 

BTW, where does "144" come from?

 

[goldsdad]

144dB because 24-bits @ 6dB/bit.

 

Have you tried 'window size: 4096' as I suggested?

It'll give a more precise (and nicer looking, IMO) spectrogram than the one you just posted.

 

P.S. Bruce repeatedly mentions an Audacity spectrogram bug in his forum, although it was fixed some time ago, AFAIK.

 

P.P.S. I just discovered Audacity 1.3.13-beta is now available.

 

 

[Jud]

wgscott writes:

 

But, then, why make it 96kHz?

 

"The benefits of 96 kHz sampling rate formats for those who cannot hear above 20 kHz"

 

http://www.nanophon.com/audio/antialia.pdf

 

[ron spencer]

Is there another program (Audition or Soundbooth) that could also be used to test these files?

 

[JR_Audio]

With Audition, you can also analyze those data and I get the same results as with Audacity (*1).

 

*1) Audacity is only wrong with ALAC files, not matter if 16 Bit or 24 Bit, but when you are using Wave, AIFF or FLAC, (not matter if they are 16 Bit or 24 Bit) then everything is correct and you get the same results as with other correct working programs.

 

The only difference in the visual results are coming mainly from different settings like upper and lower voltage or different time windows settings and lengths.

 

BTW: Roughly 144 dB is only valid for 24 Bit non FFT noise measurements (or with signals). If you are doing FFT, then the FFT window correction has to be involved at the noise floor at that could make approximately 30 dB difference with random noise.

 

Juergen

 

 

[goldsdad]

Juergen, thanks for posting.

 

"Audacity is only wrong with ALAC files"

 

Are you sure Audacity is still wrong with ALAC spectrograms? I seem to remember reading in CA forum that it had been fixed some time ago, at least in 1.3.12-beta if not before.

 

"BTW: Roughly 144 dB is only valid for 24 Bit non FFT noise measurements (or with signals). If you are doing FFT, then the FFT window correction has to be involved at the noise floor at that could make approximately 30 dB difference with random noise."

 

Are there particular settings in Audacity that you can recommend for increasing the accuracy of our spectrograms of 24-bit tracks?

 

 

 

[wgscott]

It is kind of interesting that apart from the Moody Blues, the signal above 22 kHz is robust enough to be revealed even when the dynamic range was at 80 rather than 144. I wonder why the Moody Blues recording is such an outlier?

 

[JR_Audio]

I just recently tested Audacity 1.3.13 beta on Mac OSX 10.6.7 and 16 Bit and 24 Bit ALAC are not correct. I haven't tested it under Windows, because here I use mainly Adobe Audition.

 

For the spectrogram in Audacity, I do only some basic “measurements” and have the volume range from 0 dBFS down to – 96 dBFS with 256 FFT and Hanning window. Thats's sufficient for a start.

 

Otherwise, under Mac I am running SprectraFoo X, or take the files over to Windows and measure it with Audition, where I have a running FFT (4096, Blackman-Harris) while seeing the wave file.

 

Depending of the source file, here I go down to 150 dBFS or sometimes down to 180 dBFS, but 96 dBFS is a good start, otherwise you need more knowledge about noise shaping or DSD64 noise, etc.

 

Juergen

 

[wgscott]

" 16 Bit and 24 Bit ALAC are not correct"

 

What does this mean? What isn't correct? I've used FLAC and ALAC and the results appear to be identical. What should I look for?