Spectral View Spectral Plot Range

Spectral View in CoolEdit defaults to a Logarithmic Range of 120dB. My assumption about that was that it would display signal/noise down to -120dB. Looking at an 16 bit LP recording in Spectral View, I began to question my understanding. The 16 bit dynamic range calculates to just over 96dB.  Is there really noise 120dB down?

If I set the Range to 96dB a quite noticeable amount disappears from the display. If I set it to 70dB, being generous for an LP's potential  dynamic range, much more disappears. Is it really showing me just what is above -70dB?

If this is so, is there an easy way to treat the audio itself to silence everything below a certain level, without having any large effect on the audio above that level? I would guess this should be done with some kind of compressor setting or gate, but so far I have not seen the useful part of the help files. If this is possible, could someone provide instructions, just so I can play with it?

Looking at an 16 bit LP recording in Spectral View, I began to question my understanding. The 16 bit dynamic range calculates to just over 96dB.  Is there really noise 120dB down?

You are quite correct to question your understanding!

The overall noise floor for a 16-bit recording is indeed going to be at about -96dB - when you quote a single figure for the in-band noise, generally taken as being between 20Hz-20kHz. But if you display this noise in spectral view, you are displaying and measuring the noise in individual 'channels' determined by the FFT size you've set. All the noise from each FFT channel across the band is dB-added together (that's not quite a simple sum), and the result is the overall noise figure - inevitably higher than the figure for any single FFT measurement channel.

If you want to see the difference, repeat the spectral scan with the FFT size set to different amounts. The smaller the FFT size, the larger the chunk of spectrum it is integrating, and correspondingly, the higher the noise figure will look in the display. But none of that noise seems to occur at -96dB - basically because in any one measurement channel, it isn't!

You could argue quite correctly that when you state a noise figure, you should always state the measurement bandwidth used to determine it. It is taken as read though, that when you quote a single figure and it's audio-related, that the bandwidth is 20Hz-20kHz - this has been the standard almost for ever.

The Frequency Analysis graph uses an FFT size that fits your description, but there is no FFT Size parameter in CE2K's Spectral View settings. There is a Resolution settings in "bands". I normally use 256. Either the program, or my system, can not handle anything above 2048 bands. At 4096 or above I get only a solid black display.

to modify the audio itself, I tried a Dynamics Processing settings of
Flat 1:1 above  -70dB
Expand 149.5:1 below -70dB

This seems to change the display not at all, nor what I hear. Statistics change by only a few tenths of a dB. I suppose that could mean there is nothing below -70dB that I can see or hear, but if so, why does the Spectral View display change so much when the Range is set to 70dB?

If this is so, is there an easy way to treat the audio itself to silence everything below a certain level, without having any large effect on the audio above that level?

A 16-bit signal that has been correctly dithered contains significant audio well below the noise level; just chopping that off, if it were possible, would be equivalent to or worse than not dithering in the first place.

Paul

It is quite likely there is no signal below -70dB, but regardless, observing the results is the point of my inquiry. Does your statement mean to suggest it is not possible to do what I asked?

Although I answered, I'm not actually clear what you are proposing to do.

Paul

It is quite likely there is no signal below -70dB

Actually, it's quite unlikely. To have no signal below -70dB would mean that whatever was recorded had a limited dynamic range with no natural decay, was recorded in virtual silence and anechoic conditions. How likely is that?

If you have a signal with a noise floor below -70dB, and you apply a simple compression slope to it at -70dB, then on a decent monitoring system you will get noise pumping artefacts - simply because you will have compressed the noise without covering your tracks.

With most real-world recordings, the noise floor is certainly likely to be above the -96dB theoretical noise floor of a 16-bit digitised signal, but as pwhodges suggests, there are still likely to be decaying signals, even in the noise. If you remove the dither, then you will get an abrupt cutoff when the signal actually passes through the least significant bit level, and this sounds way worse than just leaving the noise.

It's all these possibilities that make it very hard, on occasions, to set up NR to still be effective but not introduce pumping through spectral decay. It's often better to settle for less NR with no apparent artefacts, and just have done with it. But whatever you do, compression of low-level noise is really a no-no - even with a multi-band compressor, you are still likely to run into problems.

Let us start again. Spectral View settings, under Spectral Plot Style, defaults to
Logarithmic Energy Plot
120 dB Range
If your default setting was something different, it doesn’t matter, just pretend.

I always thought that the 120dB range meant that the display put a representation of any audio that is -120dBfs or higher on the screen in an appropriate color and intensity. If the range is changed, say to 150dB, audio down to -150dB would be displayed. If the range is set at 50, nothing below -50dB will be shown on screen.

Maybe that is what happens. Is it? is my first question, asked because it seems inconsistent with something else and thus a question was raised in my mind.

I think that Transform/Amplitude/Dynamics Processing does do more or less what I asked how to do; perhaps it should not be called compression. If I set
flat 1:1 above -70dB
exp infinite below -70dB
I think it does not effect anything above -70dB and greatly reduces the level of everything below -70dB, taking it down below normal audibility, anyway. I would be happy for any explanation that is (correctly) different.

Now, if I take some audio and apply this dynamic processing, especially if I take a higher threshold, say -30dB, the resulting Spectral View (Range set at 120dB) is significantly altered by the processing. This is reasonable, the audio is different, the display should reflect that.

However, go back to the original, unprocessed audio. Set the Spectral View Range to the same value as we used for processing (i.e. 70dB or 30dB). The change in the display is very different than the change made by processing the audio. Therefore, I believe, this Range parameter must mean something different, and I am back to my first question.

The second question, restated somewhat, is: How can I process the audio so that, under the default Spectral View Range (120dB), it will look the same as the unprocessed audio looks under the reduced Range Spectral View? I care not in the least whether it sounds good or bad, I just want to understand what is going on.

******************
I don’t want to have an argument about this, especially as it is irrelevant to my inquiry, but since the issue has been raised, perhaps I need some more education. It has been often claimed that 60dB is pushing the dynamic range of an LP. There are others who argue that the true dynamic range is greater than possible on CD, but that isn’t the issue. Let us just suppose, for the purpose of going forward, that 60dB is a good real life figure.

If the dynamic range is 60dB, and the audio is normalized to 0dB, does that not mean that no music, voice, whatever is in the recording, quieter than -60dB can be heard? There is noise aplenty below -60dB, but no signal can be separated from it, yes?

There were probably lower level sounds in the performance from which the recording was made. Since this was for LP, it was probably on tape and the tape probably captured something below -60dB. However, by the time it is mastered for LP, scratched onto a master disk, pressed into plastic, scratched off the LP, and put out your speakers, you no longer have any of that lower level detail (unless it was compressed to a higher level during mastering, but then it isn’t below -60dB anymore, so it doesn’t count).

Is this a misunderstanding? If not, is it not reasonable to say there is no signal below -70dB?
If you keep 10dB of noise below the lowest signal (-60dB), but nothing lower than that, how likely are you to have something that sounds funny?

I always thought that the 120dB range meant that the display put a representation of any audio that is -120dBfs or higher on the screen in an appropriate color and intensity.

Yes.  But that -120 is not on the same scale as the full-bandwidth level figures you use elsewhere; but this has been said.   Consider - if you have white noise at -70dBfs, and you then band-limit this to a small fraction of an octave, the level of this fragment will be much lower than the original -70 - this is what you are seeing in the spectral plot, which is a plot of all those fragments across the full audio frequency range.

However, go back to the original, unprocessed audio. Set the Spectral View Range to the same value as we used for processing (i.e. 70dB or 30dB). The change in the display is very different than the change made by processing the audio. Therefore, I believe, this Range parameter must mean something different, and I am back to my first question.

They are unrelated - one is a display parameter, the other is a parameter for some processing.

The second question, restated somewhat, is: How can I process the audio so that, under the default Spectral View Range (120dB), it will look the same as the unprocessed audio looks under the reduced Range Spectral View?

I suppose you would be applying the dynamics processing from 0dB downwards, with both attack and recovery times of 0.  Sort of.  Of course, really the two spectral views (with different ranges) look the same - you have simply stretched them differently across the colour scale.  It is quite hard to talk about this comparison, as it truly makes no sense anyway.

If the dynamic range is 60dB, and the audio is normalized to 0dB, does that not mean that no music, voice, whatever is in the recording, quieter than -60dB can be heard?

No, it doesn't - you can hear stuff within noise, and it is often stuff that matters (reverb tails, etc).

If you keep 10dB of noise below the lowest signal (-60dB), but nothing lower than that, how likely are you to have something that sounds funny?

That's just noise gating, isn't it?  Try it and see if it suits your circumstances.

Paul

I had to go over my reasoning a number of times before I understood the problem with my view: The dynamics process is strictly time domain. At any given instant, all it sees is one factor, the total signal amplitude. It makes its yes/no decision based only on that. No doubt you were essentially saying that, but the emphasis was on the spectral aspects, which was not what I was missing.

As a practical matter, I can understand how a sharp lower cutoff can make a big difference. Raise the threshold (-60db, -50dB, ...) and it becomes more and more obvious as more and more of the audio just drops out. However, for the LP recording I’ve been playing with, I have not been able to find any place where a -70dB cutoff is audible, at least on headphones. Even -60dB makes little difference. This is a old mono recording of medieval music, made in a very resonate space. It has a relatively wide dynamic range and sounds quite good to me.

Doing the dynamics processing was not an exercise for practical purposes. I don’t see much use for it, except perhaps as part of a “loudness” compression to make music to be played in a very noisy environment, such as an automobile. Here, I just wanted to find out the sound of what I see on screen when reducing the Spectral View range.

From what you are saying, I don't for one moment think that you've grasped all the consequences of this at all, but never mind...

The dynamics process is strictly time domain. At any given instant, all it sees is one factor, the total signal amplitude. It makes its yes/no decision based only on that.

Not true, I'm afraid. The detector does have a distinct rise-time, and even with lookahead the end results aren't instantaneous either, because of the interaction between the risetime of the signal and what's controlling its final amplitude - which can screw transients pretty badly.

The general argument is that if, by any means, you introduce any amplitude non-linearity into a signal, the chances are that it's going to be audible - more so under some circumstances than others. In theory, using noise reduction on a signal doesn't alter the dynamic range, although the artefacts it introduces sometimes sound worse than what you are trying to get rid of. But in general if it's applied with caution, it's preferable to the relatively crude results you get from dynamic expansion (aka gating) - which is what you were describing earlier.

I didn’t define “instant,” and maybe it is a poor choice to describe the process, but the point is that only amplitude and time are involved. Frequency of the audio doesn’t enter into consideration. The process doesn’t know anything about the frequency distribution of the signal.

I didn’t define “instant,” and maybe it is a poor choice to describe the process, but the point is that only amplitude and time are involved. Frequency of the audio doesn’t enter into consideration. The process doesn’t know anything about the frequency distribution of the signal.

That's why I said that you didn't yet understand it all. The process doesn't 'know' anything - but if you think that time and frequency aren't related in this context, you'd be fundamentally mistaken. The relationship between amplitude and time is not an instantaneous one at all, but related to rates of change. And if the rate of change of output isn't as fast as the rate of change of higher frequencies and transients, then they get smeared, and significantly alter the characteristics of the output.

[all]

Sure, but that is why it might not sound right, not why I could not get processed audio that looked, in Spectral View, like unprocessed audio with a reduced Range value. For Dynamic Processing, only the signal amplitude is relevant, from this point of view, . No doubt I don't understand everything about compression, expansion, and related topics. Yes, I know that all the information that can be accessed by any process is inherent in the data. My post is about one particular thing, not about the whole universe of audio.

My post is about one particular thing, not about the whole universe of audio.

Yes, but your 'one thing' is inter-related with everything else, so you can't simply ignore that just because it's inconvenient...