digital theory versus reality

[Joe Bloggs]

... and when all the bellyaching about possible DAC, amp and file format gremlins is said and done, the "reality" is I can still stick a $50 pair of headphones into a $100 DAC/amp and get superior musical performance playing redbook CDs to TOTL rigs costing tens of thousands per part playing DSD512 or whatever, because I bothered to characterize the headphone's frequency response on my ears and my own loudspeaker-to-headphone HRTF and compensated for both using (shock horror) parametric equalizers and cross-channel convolution DSPs :rolleyes:

Yes, my central question is what do we need to know to reproduce music with fidelity. But you are leaving an important concept undefined here. What is the definition of "superior musical performance"? 

There are ABX tests for determining whether sufficiently similar stimuli are differentiable at all by the subject, then there are preference tests to determine which of a set of quite different stimuli is subjectively most preferred by the subject. Since no HiFi audio reproduction comes close enough to the real thing to require ABX testing, and since the effect of correcting a pair of headphones' frequency response and HRTF for oneself are anything but subtle compared to "plain" playback, we should be investigating via preference testing.

https://en.wikipedia.org/wiki/Preference_test

Still it should not be hard to demonstrate that the effect of the effects I mentioned (EQ and HRTF, and a new item, room treatment and correction for loudspeaker systems) on correcting the perceived signal towards ideal stereo playback are disproportionately larger than any "fidelity"-based incremental improvements on garden-variety DACs, amps and even headphones and loudspeakers--because DACs and amps are at the mercy of the distortions introduced by headphones and loudspeakers, loudspeakers are at the mercy of whatever the room does to the sound they make (however perfect) while headphones are simply physically unable to pretend to be loudspeakers. The effects I mentioned directly manipulate the source signal to nullify the bad effects mentioned above, which does more for ultimate fidelity than any misguided attempt to preserve the signal as it is through every stage in the playback chain--if the signal is not customized ("pre-distorted") to the playback system, it will be distorted by the loudspeaker / headphone / room beyond recognition, if you want any semblance to the desired original signal reaching your ears you need to predistort the digital signal beyond recognition to combat the said distortions that will occur down the chain. This is something that 99% of audiophiles don't understand.

 

[johncarm]

There are ABX tests for determining whether sufficiently similar stimuli are differentiable at all by the subject, then there are preference tests to determine which of a set of quite different stimuli is subjectively most preferred by the subject. Since no HiFi audio reproduction comes close enough to the real thing to require ABX testing, and since the effect of correcting a pair of headphones' frequency response and HRTF for oneself are anything but subtle compared to "plain" playback, we should be investigating via preference testing.

https://en.wikipedia.org/wiki/Preference_tes

 
Are you saying there's no point to investigating how much a reproduction sounds like the real thing? I.e. say we are in the concert hall during a recording of an orchestra, and we get a sense of the sound. Then we go into the control studio and check the recording. Are you saying there is no point to asking whether they are similar?

 

[Joe Bloggs]

See my edited post.

 

[johncarm]

See my edited post.

Okay, so I see you saying that the reproduced sound is different in many ways than the original. But I'm not sure that answers my question.
 
Musicians ask all the time in what ways the reproduced sound is similar to the original. For instance, a conductor may be adjusting articulation and balance in order to achieve a certain effect. He goes into the control room and simply asks, is the effect clearly audible?
 
So are you saying this is pointless?

 

[gregorio]

  [1] Yes, my central question is what do we need to know to reproduce music with fidelity. [2] What is the definition of "superior musical performance"? 

 
1. That depends on what you mean by music. If you mean acoustic sound waves, then as I've said repeatedly, amplitude and frequency. If you think there is something else, please answer the question and state what!!
 
2. "Superior musical performance" is subjective and therefore there is no one definition.
 
This thread is effectively now converging with your other thread!
 
G

 

[johncarm]

   
1. That depends on what you mean by music. If you mean acoustic sound waves, then as I've said repeatedly, amplitude and frequency. If you think there is something else, please answer the question and state what!!
 
2. "Superior musical performance" is subjective and therefore there is no one definition.
 
This thread is effectively now converging with your other thread!
 
G

 
If the mod wants to merge the threads that's fine with me. I didn't mean to distract the forum from people who don't care about this topic.
 
[1] Your language is vague here. It sounds like you are saying "All we need to know to reproduce acoustic sound waves is amplitude and frequency." I can only guess what you are trying to say. Maybe you are saying that any audio device can be characterized by its transfer function?
 
[2] I asked Joe what his definition is. 

 

[SoundAndMotion]

If you mean acoustic sound waves, then as I've said repeatedly, amplitude and frequency. If you think there is something else, please answer the question and state what!!

The problem is that that has been incorrect… repeatedly, not only in this thread, but in others, for quite a while. Maybe we can figure out where the problem is.

You started with this early in the thread:

I see where you are coming from. As a musician we have 3 fundamental basics: Volume, pitch and timing. If we're taught anything at all about the science of sound, we're usually taught that volume (loudness) = amplitude and pitch = frequency and as amplitude and frequency is all we are able to record or reproduce, this leaves timing rather unaccounted for! Unfortunately, "volume = amplitude" and "pitch = frequency" is not actually true, it's just a useful over-simplification which is only partially true or true sometimes. Without going into reams of details, our timing component is already dealt with by frequency, because frequency is the number of audio cycles per second (Hertz), so timing is implicit in measuring/recording/reproducing frequency. In other words, a spike (transient) is in fact defined only by it's amplitude and frequency content. This fact invalidates most of the rest of your post.

This starts off clear and makes sense as to how a musician sees it. And I understand why you say elsewhere that loudness=amplitude is an oversimplification of a complex process, but we know that if all else is equal (duration, waveform, etc.) that loudness is monotonically, but not linearly, related to amplitude. So they are related. But what is wrong with pitch=frequency? And what happens to timing? You know duration, ADSR and all that? Saying that frequency is “cycles per second (Hertz), so timing is implicit in measuring/recording/reproducing frequency” is flat wrong.

I’m guessing you have been thrown off by a partial understanding of the frequency domain data (the Fourier transform of the time domain data) for a recorded sound (or any type of data).
In the time domain, we have the real-valued amplitude (sound wave or voltage or….) as a function of time. In this domain, we don’t need to worry about frequency; it is in the data. A microphone doesn’t transduce frequency, rather it transduces the sound wave with time. The voltage fluctuates along with the fluctuating sound wave. There is no voltage or signal that tells us the frequency.
In the frequency domain, we have the complex-valued amplitude as a function of frequency. The complex value gives us magnitude AND phase. The phase being critical to understanding the musicians concept of “timing”.
We can easily see this in two ways: To fully define a sine wave, you need magnitude, frequency AND phase.
Also, notice that the ONLY difference between an impulse and white noise is the phase. The magnitude versus frequency spectrum can be completely the same (i.e. flat for all frequencies).

If I have misunderstood what you have written, please clarify. i don’t think anyone needs “reams of details” to understand why we don’t agree.

 

[Joe Bloggs]

Also, notice that the ONLY difference between an impulse and white noise is the phase. The magnitude versus frequency spectrum can be completely the same (i.e. flat for all frequencies).

Show me a white noise spectrum with *identical* amplitude over all frequencies (as it is with an ideal impulse) and I'll show you a flying pig. :rolleyes:

Of course you're right that a fourier analysis of a waveform involves frequency, magnitude AND phase, and that extreme distortions in the latter over a long time window (you're talking about a single FT of a whole audio clip) can be easily audible. But the fact remains that phase distortions in audio are one of the things in audio we are less sensitive to.

 

[SoundAndMotion]

Show me a white noise spectrum with *identical* amplitude over all frequencies (as it is with an ideal impulse) and I'll show you a flying pig. :rolleyes:

Happy to. Do you just want a graph or better yet the data? Where should I send it? By the way, how do you define "white noise"?

As for the flying pig, I'd like to see a real one, not Photoshop or just one thrown in the air. :roll eyes:

Also, I made no mention of phase distortion, only that phase cannot be left out of a complete description.

 

[Joe Bloggs]

Show me a white noise spectrum with *identical* amplitude over all frequencies (as it is with an ideal impulse) and I'll show you a flying pig. :rolleyes:

Happy to. Do you just want a graph or better yet the data? Where should I send it? By the way, how do you define "white noise"?

As for the flying pig, I'd like to see a real one, not Photoshop or just one thrown in the air. :roll eyes:

Also, I made no mention of phase distortion, only that phase cannot be left out of a complete description.

I'm not aware of any definition of white noise that allows you to produce a flat FT amplitude spectrum except in the way one might assemble a Boeing 747 by throwing its component parts into a typhoon 10 billion times and let chance put one together for you sometime.

This is what a typical white noise spectrum looks like. Flat on average to be sure, but nowhere near a straight flatline like one obtains from an impulse.


My flying pig can only be as real as whatever "flat FT spectrum" "white noise" you show me turns out to be.

 

[spruce music]

Joe Blogs beat me to it.  Was about to post the same thing.
 
With a good impulse you will get the same level in each FFT bin which will graph as a flat straight line vs the jagged one in his posted FFT.

 

[SoundAndMotion]

I don't accept his version as the only possible one. And I don't understand the gibberish about 747's, but the Wikipedia article on white noise is pretty clear for me. First sentence:
"In signal processing, white noise is a random signal with a constant power spectral density."

Again, do you just want graphs? That is just "noisy" in the time domain and a ruler-flat spectrum. Would you believe it? Can I upload or send you a .wav for you to test yourself? Piece of cake, but I'll need a few minutes. Do you want to specify length or amplitude or anything? I'll get started and give it to you when I know what you want.

 

[Joe Bloggs]

I don't accept his version as the only possible one. And I don't understand the gibberish about 747's, but the Wikipedia article on white noise its pretty clear for me. First sentence:
"In signal processing, white noise is a random signal with a constant power spectral density."

Again, do you just want graphs? That is just "noisy" in the time domain and a ruler-flat spectrum. Would you believe it? Can I upload or send you a .wav for you to test yourself? Piece of cake, but I'll need a few minutes. Do you want to specify length or amplitude or anything? I'll get started and give it to you when I know what you want.

Oh, it'd be trivial to fire up matlab and insert a row of 1's for the amplitude and a bunch of random numbers for the phase angle and then convert that into a waveform. I'm just not aware of any definition of "white noise" for which such a waveform qualifies.

I could say I'm looking to analyse whatever waveform you produce using a 65536 sample rectangular window fourier transform, so you just need to produce 65536 samples. But really, any sample that produces a ruler flat FR, I wouldn't call it a sample of white noise unless you can explain how you went about obtaining said sample from a random process.

 

[SoundAndMotion]

It would be trivial to:
1. create an N-element array of 1's for the "r" in polar notation
2. create an N-element array of random numbers between 0 and 2*pi or between -pi and pi for the theta in polar notation.
3. convert the array to real and complex values.
4. create a copy, reverse the order, and create the complex conjugate
5. make the larger array [1+0i] [first complex array] [1+0i] [second reversed array]
6. do an inverse FFT

I'm not aware of any definition of white noise for which this doesn't qualify. Read the link I gave for the Wikipedia article. Please note that I don't claim this is the only way AND that my original post here said "The magnitude versus frequency spectrum can be completely the same". Also note that although this is a synthetic dataset, so is an ideal impulse!

The phase is a random process.

I'm waiting for the pig...

 

[Joe Bloggs]

It would be trivial to:
1. create an N-element array of 1's for the "r" in polar notation
2. create an N-element array of random numbers between 0 and 2*pi or between -pi and pi for the theta in polar notation.
3. convert the array to real and complex values.
4. create a copy, reverse the order, and create the complex conjugate
5. make the larger array [1+0i] [first complex array] [1+0i] [second reversed array]
6. do an inverse FFT

I'm not aware of any definition of white noise for which this doesn't qualify. Read the link I gave for the Wikipedia article. Please note that I don't claim this is the only way AND that my original post here said "The magnitude versus frequency spectrum can be completely the same". Also note that although this is a synthetic dataset, so is an ideal impulse!

That would only be random in the phase domain. It is completely deterministic in the amplitude domain. And when the goal of the contest is to have a ruler flat FR, it's like saying you can have any random person build an audio amplifier that measures within 0.1dB from 20Hz to 20kHz, then saying you win because you added the condition that the random person is actually a random "audio electronics engineer with at least 10 years experience".

There is no requirement for an ideal impulse to be a randomly generated entity, unlike for white noise.

It's not fair but it's the terms you agreed to when entering the flying pig contest

Finally, if you did take the bother to do all this, I suggest you make a 64-sample block of this so-called "noise" and compare it in listening with an ideal impulse. You might be surprised how similar they sound :rolleyes: