• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!

Time vs. Frequency Domain

MRC01

Major Contributor
Joined
Feb 5, 2019
Messages
3,422
Likes
4,029
Location
Pacific Northwest
Reading about perception in the time vs. frequency domains, I came across this study from several years ago:
https://phys.org/news/2013-02-human-fourier-uncertainty-principle.html

It's not an Earth shattering revelation that human hearing acuity is different in the time and frequency domains. I wonder if it is different enough to be relevant from an audio engineering perspective. That is, to what extent (if any) is audio engineering based on a notion of symmetric acuity? Or maybe the difference, if any, is too small to matter.

I'm trying to imagine a simple experiment that could test this asymmetry of acuity in time vs. frequency domains. How about this: take an audio signal with strong HF energy extending to at least 20 kHz, with transient impulses (not just a mix of pure tones). The signal could be natural or an artificial test signal. For each listener, separately determine the highest frequency pure tone he can perceive at a reasonable level. Call this frequency Fl. Take the discrete FT of that audio signal, from that FT remove all frequencies above Fl. Reconstruct the signal from the redacted DFT. See if the the listener can differentiate it from the original non-filtered signal in a DBT.

The idea: if our hearing acuity is asymmetric in the time & frequency domains, it might be possible to detect the absence of frequencies we can't hear as pure tones, if removing them changes the signal in the time domain (smearing or spreading impulse timing). Or not?
 
Last edited:

SIY

Grand Contributor
Technical Expert
Joined
Apr 6, 2018
Messages
10,383
Likes
24,749
Location
Alfred, NY
You might consider some literature searches.
 
OP
MRC01

MRC01

Major Contributor
Joined
Feb 5, 2019
Messages
3,422
Likes
4,029
Location
Pacific Northwest
Yeah, I'm trying... that's how I found the article above. I'm looking for more, open to suggestions.
It seems that not everyone believes they proved what they think they proved: https://arxiv.org/abs/1501.06890
 
Last edited:

SIY

Grand Contributor
Technical Expert
Joined
Apr 6, 2018
Messages
10,383
Likes
24,749
Location
Alfred, NY
The Fourier uncertainty paper is not really related to what I think you're asking.

You might start with Nishiguchi et al "Perceptual Discrimination between Musical Sounds with and without Very High Frequency Components."
 

andreasmaaan

Master Contributor
Forum Donor
Joined
Jun 19, 2018
Messages
6,652
Likes
9,399
The Fourier uncertainty paper is not really related to what I think you're asking.

You might start with Nishiguchi et al "Perceptual Discrimination between Musical Sounds with and without Very High Frequency Components."

Not sure why, but also available freely here.
 
OP
MRC01

MRC01

Major Contributor
Joined
Feb 5, 2019
Messages
3,422
Likes
4,029
Location
Pacific Northwest
Thanks for the reference! I've downloaded the latter paper. I haven't read it yet, but my first thought is that if improper antialias filters allow frequencies above Nyquist to leak through during A-D, then a theoretically perfect D-A reconstruction filter based on Shannon-Whittaker sinc(t) would not match the original analog waveform. That is, it would reconstruct the "perfect" properly bandwidth-limited analog wave that produced those digital samples. But the original analog wave is different from this wave because it wasn't properly bandwidth limited (since freqs above Nyquist leaked in and got aliased into the passband). If the testers allow this to happen, test subjects might differentiate the absence of these supersonic sounds, creating positives that are false because they're not really hearing those sounds, but only hearing lower frequency passband distortion from improper filtering.

As I read the paper, no doubt I'll find they accounted for this and other things.
 
OP
MRC01

MRC01

Major Contributor
Joined
Feb 5, 2019
Messages
3,422
Likes
4,029
Location
Pacific Northwest
I wouldn't be surprised, because in the hypothetical flawed test example I gave, it would only require hearing their aliased artifacts in the passband, left over from improper AA filtering when the recording was created.

However, the paper that SIY referenced used a different test method, and 2 of the listeners detected the presence or absence of supersonic material better than random guessing. That surprised me.
 

andreasmaaan

Master Contributor
Forum Donor
Joined
Jun 19, 2018
Messages
6,652
Likes
9,399
Me too. Best explanation I can think of is that that 25dB of audio band IMD the super tweeter created poked through the masking threshold at a couple of points in playback of the 2nd stimulus.

Anyone got any other theories?
 

Blumlein 88

Grand Contributor
Forum Donor
Joined
Feb 23, 2016
Messages
20,521
Likes
37,050
I wouldn't be surprised, because in the hypothetical flawed test example I gave, it would only require hearing their aliased artifacts in the passband, left over from improper AA filtering when the recording was created.

However, the paper that SIY referenced used a different test method, and 2 of the listeners detected the presence or absence of supersonic material better than random guessing. That surprised me.
Some 2 % or so of young adults have some sensation of sound to 22 khz or so. That at very high SPL. Those are also potentially sensitive to effects of filters at the edge of this range if the transition zone is short. This paper used a set of filters at 21 khz with only a 1 khz transition band. Around 4 khz would have been better, but you cause other issues with the test done this way.
 
Last edited:
OP
MRC01

MRC01

Major Contributor
Joined
Feb 5, 2019
Messages
3,422
Likes
4,029
Location
Pacific Northwest
I'd like to see a test where the high frequencies turned on or off were customized to each listener, just above his hearing threshold for pure tones. And with the audio signal having strong impulses or transients. The goal: test whether acuity in the time domain is the same or different from acuity in the frequency domain.
 

andreasmaaan

Master Contributor
Forum Donor
Joined
Jun 19, 2018
Messages
6,652
Likes
9,399
Some 2 % or so of young adults have some sensation of sound to 22 khz or so. That at very high SPL. Those are also potentially sensitive to effects of filters at the edge of this range if the transition zone is short. This paper used a set of filters at 21 khz with only a 1 khz transition band. Around 4 khz would have been better, but you cause other issues with the test done this way.

The two subjects who "passed" the test had their hearing acuity tested and these were the results:
1550002966687.png


Here is the filter used:
1550003044430.png


Transition band effects could conceivably account for Subject 2's ability to discriminate, but it's hard to imagine how it could account for Subject 9, I would have thought?

Another line of thought: although the subjects were found not to be able to hear at 22KHz/90dB SPL, the stimuli had a continuous SPL of close to 90dB:
1550003316227.png


Some peaks would presumably have risen above 90dB. Perhaps these were heard.
 

solderdude

Grand Contributor
Joined
Jul 21, 2018
Messages
15,891
Likes
35,912
Location
The Neitherlands
impulses and transients in music are in the 3-7kHz range.
People tend to believe it is in the 10-20kHz range but alas there isn't much energy there.
Things like 'sparkle' and realism in cymbals is found in the upper frequencies.
 
OP
MRC01

MRC01

Major Contributor
Joined
Feb 5, 2019
Messages
3,422
Likes
4,029
Location
Pacific Northwest
impulses and transients in music are in the 3-7kHz range.
People tend to believe it is in the 10-20kHz range but alas there isn't much energy there.
Things like 'sparkle' and realism in cymbals is found in the upper frequencies.
It's generally true that the frequency energy of most natural sounds drops about 6 dB per octave from the midrange on up. By the time you get to 15 kHz there's not much left. But there are exceptions. A close-miced recording of castanets or other small percussive sounds using wide bandwidth mics has energy well above 10 kHz. I just recorded lightly jangling my keys in front of my mic (an old recording engineer trick I read about long ago), and the spectrum is bimodal with equal amplitude peaks at 2 kHz and 20 kHz, more energy at 25 kHz than 1 kHz. That's with a Rode NT1A mic, which is already rolling off before reaching 20 kHz so the actual HF energy of the sound is even more than what I captured.

When I listen to this and apply low-pass filters from 10 kHz on up, my subjective perception is at first the filter affects mainly timbre, but as the filter frequency increases and approaches my upper hearing limit, the timbre differences fade away and I perceive it more as timing, smearing the transients so they are less "crisp".

This is why I'm curious about hearing acuity in general in the frequency vs. time domains.
 

andreasmaaan

Master Contributor
Forum Donor
Joined
Jun 19, 2018
Messages
6,652
Likes
9,399
When I listen to this and apply low-pass filters from 10 kHz on up, my subjective perception is at first the filter affects mainly timbre, but as the filter frequency increases and approaches my upper hearing limit, the timbre differences fade away and I perceive it more as timing, smearing the transients so they are less "crisp".

I agree with you about the frequency response, but I'd be sceptical if I were you of the perceived loss of timing, smearing, lack of crispness etc, given that you're specifically listening for this. Why not try using an ABX comparator, testing the original recording against a version that's high-pass filtered (linear phase) at a frequency such that the transition band doesn't begin until just above the limit of your hearing (at whatever decibel level you perform the test at)?
 
OP
MRC01

MRC01

Major Contributor
Joined
Feb 5, 2019
Messages
3,422
Likes
4,029
Location
Pacific Northwest
You mean low-pass filtered? Yep, that's the experiment I've done. Though I used Audacity's parametric and low-pass filters, and I don't know how they're implemented: minimum or linear phase.

During the ABX test, I was listening for any difference at all without trying to describe it. My description of it shifting from timbre to timing as it approached my hearing threshold is purely incidental and subjective. As such that has no objective or experimental meaning.
 

andreasmaaan

Master Contributor
Forum Donor
Joined
Jun 19, 2018
Messages
6,652
Likes
9,399
You mean low-pass filtered? Yep, that's the experiment I've done. Though I used Audacity's parametric and low-pass filters, and I don't know how they're implemented: minimum or linear phase.

During the ABX test, I was listening for any difference at all without trying to describe it. My description of it shifting from timbre to timing as it approached my hearing threshold is purely incidental and subjective. As such that has no objective or experimental meaning.

Ok, I seem to recall that Audacity's filters require some work to set up correctly. But besides that, what I was suggesting was that you compare the filtered and unfiltered signals using something like this to make sure that you're not hearing the difference only because you're listening for it.

If you do that and it confirms that the difference you're hearing is real, then the next step would be to look at your Audacity filter settings to make sure the filter is not introducing artefacts below the upper frequency threshold of your hearing.

Another couple of things to check would be how you determined the upper frequency threshold of your hearing. And was it on the same system that you're comparing the current signals on? Finally, you'd want to rule out IM distortion generated by your system within the audio band (a near impossible task TBH - something that the authors of the study took great care with and still couldn't quite eliminate).

Of course there's no need to do all this :) Just suggesting some ways you could begin to make your test more rigorous.

PS. when I said "listening for" time smear etc., what I meant was that you are listening for a difference (not the specific differences you're hearing necessarily).
 
OP
MRC01

MRC01

Major Contributor
Joined
Feb 5, 2019
Messages
3,422
Likes
4,029
Location
Pacific Northwest
I'm using my own ABX app, similar to the one you linked. Got most of the bases you mentioned covered (clean sound card, headphone amp, headphones), though it's an informal experiment, the results are just for my own curiosity, not to submit to the AES. But as much as experimenting on ourselves is fun, the results of one person are irrelevant. I'm more interested in results across a population.

I made another cleaner recording of keys jangling. The energy is focused in the top octave, 10k - 20k. This is a good example of a natural sound that is the exception to the rule of decreasing energy with frequency. The mic response peaks at 12k but the recording peaks at 17k, so the actual HF energy is much stronger than captured here. Incidentally, this recording rates a DR20. I'd post the FLAC here so others can play with it, but the site won't let me so I'll tease with pictures. ;)
keyJangle.png
 

andreasmaaan

Master Contributor
Forum Donor
Joined
Jun 19, 2018
Messages
6,652
Likes
9,399
I'm using my own ABX app, similar to the one you linked. Got most of the bases you mentioned covered (clean sound card, headphone amp, headphones), though it's an informal experiment, the results are just for my own curiosity, not to submit to the AES. But as much as experimenting on ourselves is fun, the results of one person are irrelevant. I'm more interested in results across a population.

I made another cleaner recording of keys jangling. The energy is focused in the top octave, 10k - 20k. This is a good example of a natural sound that is the exception to the rule of decreasing energy with frequency. The mic response peaks at 12k but the recording peaks at 17k, so the actual HF energy is much stronger than captured here. Incidentally, this recording rates a DR20. I'd post the FLAC here so others can play with it, but the site won't let me so I'll tease with pictures. ;)
View attachment 21811

Nice :) Sorry I hadn't realised from your previous posts that you were already doing this with an ABX comparator. Now I see that it was written right there in your previous post to mine.

What Audacity filter settings are you using?
 
Top Bottom