[pfzar]

Seriously you need a chill pill.

nothing false was stated about the FFT. You continue on your path.
Cheers

 

[csglinux]

We've done enough thread-derailing. If you've gone beyond technical discussion, it's time to add you to my list of blocked users.

To THX though - I'm surprised you'd let one of your employees publicly display this combination of ignorance and unprofessionalism under your corporate name. Even his insults aren't funny or original.

 

[jude]

We've done enough thread-derailing. If you've gone beyond technical discussion, it's time to add you to my list of blocked users.

To THX though - I'm surprised you'd let one of your employees publicly display this combination of ignorance and unprofessionalism under your corporate name. Even his insults aren't funny or original.

@csglinux, in my opinion you kind of set the tone of this conversation. I've known @pfzar for some time, and when I first started doing audio measurements he was immensely knowledgable and helpful. If you have a chance to discuss the research he and his colleagues have done (at least some of which you can find published in the AES library), I think you'll find the same about him.

I think part of the disconnect in this discussion is that you seem to have interpreted what was said in the video as an attack on FFT analysis. I thought @Mr.Jacob's mention of Dr. Genuit's soup analogy (not to mention the rest of the video discussion) was enough to make this clearer, but it seems I was wrong.

...I feel somebody here needs to defend the humble fast Fourier transform...

No, that's not what's needed here, because....

...This video bashes "FFT analysis" several dozen times...

....that's simply not the case.

...IMHO, any suggestion that an FFT analysis is completely wrong is completely wrong...

But that's not what was said. (Read the posts by @Mr.Jacob, @pfzar, @arnaud, and @castleofargh above.)

Regarding psychoacoustics, here's another very simple example (because when it comes to the subject of psychoacoustics, the only knowledge I have to draw on involves the very simplest examples):

Fig.1 (above) 40 dBSPL stepped frequency sweep

That was just an analyzer loopback, but assume for the sake of example that Fig.1 represented an acoustical capture (through a linear measurement microphone) of a stepped frequency sweep. At any of the steps in that range, we know the measured level is 40 dBSPL. However, what if the question we were trying to answer was "Would human perception of that output be best represented by this measurement (in Fig.1 above)?"

The answer is "no."

In answering "no," would one be bashing FFT analysis? Of course not.

To help us better answer that simple question, there's a commonly referenced psychoacoustic measure that we use to represent perception of equal loudness that we commonly call the Fletcher-Munson Curve (but that has been modernized with equal loudness contours in ISO 226:2003 (shown below)):

Fig.2 (above) Equal loudness contours (ISO 223:2003)

Because loudness is our brain's perception of sound pressure, then loudness is a psychoacoustic phenomenon -- so let's look at a psychoacoustic measure. Looking at the 40-phon curve in Fig.2, we can correlate the human perception of the output in Fig.1. The 40-phon curve shows us that to perceive a 20 Hz tone as equally loud as a 1 kHz tone at 40 dBSPL, that 20 Hz tone would need to be at a sound pressure level of 100 dBSPL. A 125 Hz tone would have to be 60 dBSPL to be perceived by a human as equally loud as a 1 kHz tone at 40 dBSPL. Etc.

So while Fig.1 would not represent human perception of the output (a 40 dBSPL stepped frequency sweep), perhaps an inverse of the 40-phon curve in Fig.2 (a psychoacoustic measure) would better represent human perception of that output. (@pfzar, @Mr.Jacob, @arnaud, or anyone else more qualified, please correct me if I'm wrong or have oversimplified this.)

So is the measurement in Fig.1 wrong? No, it is what it is. But what it is not is the best answer to the question "How would a human perceive the output that generated the FFT-based measurement in Fig.1?"

And that's not bashing FFT analysis.

...With the relevant phase information, the Fourier transform is completely reversible - one can hop back and forth between time and frequency domains and lose nothing that the microphone was able to capture. There *may* be some rare hypothetical pyschoacoustic effects that aren't easily captured, e.g., separate L and R FFTs without crossfeed wouldn't capture the beat frequencies generated in the brain when the left and right ear are fed separate tones, but I don't think this analysis covered anything like that, did it? It sounds like you're just using a weighting function (one geared more toward speech recognition?). That's fine (it should be obvious that overall sound pressure level is a silly metric), but an FFT analysis with a weighting function is still an FFT analysis. Or was there something else going on beyond a weighting function that wasn't mentioned in the video?...

The phase information, and hopping back and forth between time and frequency domains, still would not yield the answer to the question raised in my rather plain example above. And it could reasonably be said that the above example is a far more rudimentary psychoacoustic question than the many-faceted question asked in the video: Which of the two tested ANC headphones had more effective noise canceling as perceived by humans?

And, again, that's not bashing FFT analysis. I genuinely do not understand how the video was interpreted that way, but it seems to be largely due to your conclusion that anyone's "bashing" FFT (which is not the case), and your insistence, then, on defending it to the death from an attack that hasn't occurred.

The conversation with Jacob in the video could open up a whole lot of interesting follow-up discussion about ANC, psychoacoustics, audio measurements, etc. I think we need to move past this idea that FFT has to be defended here, as it does not.

 

[csglinux]

@csglinux, in my opinion you kind of set the tone of this conversation. I've known @pfzar for some time, and when I first started doing audio measurements he was immensely knowledgable and helpful. If you have a chance to discuss the research he and his colleagues have done (at least some of which you can find published in the AES library), I think you'll find the same about him.

I think part of the disconnect in this discussion is that you seem to have interpreted what was said in the video as an attack on FFT analysis. I thought @Mr.Jacob's mention of Dr. Genuit's soup analogy (not to mention the rest of the video discussion) was enough to make this clearer, but it seems I was wrong.



No, that's not what's needed here, because....



....that's simply not the case.



But that's not what was said. (Read the posts by @Mr.Jacob, @pfzar, @arnaud, and @castleofargh above.)

Regarding psychoacoustics, here's another very simple example (because when it comes to the subject of psychoacoustics, the only knowledge I have to draw on are the very simplest examples):


Fig.1 (above) 40 dBSPL stepped frequency sweep

That was just an analyzer loopback, but assume for the sake of example that Fig.1 represented an acoustical capture (through a linear measurement microphone) of a stepped frequency sweep. At any of the steps in that range, we know the measured level is 40 dBSPL. However, what if the question we were trying to answer was "Would human perception of that output be best represented by this measurement (in Fig.1 above)?"

The answer is "no."

In answering "no," would one be bashing FFT analysis? Of course not.

To help us better answer that simple question, there's a commonly referenced psychoacoustic measure that we use to represent perception of equal loudness that we commonly call the Fletcher-Munson Curve (but that has been modernized with equal loudness contours in ISO 226:2003 (shown below)):


Fig.2 (above) Equal loudness contours (ISO 223:2003)

Because loudness is our brain's perception of sound pressure, then loudness is a psychoacoustic phenomenon -- so let's look at a psychoacoustic measure. Looking at the 40-phon curve in Fig.2, we can correlate the human perception of the output in Fig.1. The 40-phon curve shows us that to perceive a 20 Hz tone as equally loud as a 1 kHz tone at 40 dBSPL, that 20 Hz tone would need to be at a sound pressure level of 100 dBSPL. A 125 Hz tone would have to be 60 dBSPL to be perceived by a human as equally loud as a 1 kHz tone at 40 dBSPL. Etc.

So while Fig.1 would not represent human perception of the output (a 40 dBSPL stepped frequency sweep), perhaps an inverse of the 40-phon curve in Fig.2 (a psychoacoustic measure) would better represent human perception of that output. (@pfzar, @Mr.Jacob, @arnaud, or anyone else more qualified, please correct me if I'm wrong or have oversimplified this.)

So is the measurement in Fig.1 wrong? No, it is what it is. But what it is not is the best answer to the question "How would a human perceive the output that generated the FFT-based measurement in Fig.1?"

And that's not bashing FFT analysis.



The phase information, and hopping back and forth between time and frequency domains, still would not yield the answer to the question raised in my rather plain example above. And it could reasonably be said that the above example is a far more rudimentary psychoacoustic question than the many-faceted question asked in the video: Which of the two tested ANC headphones had more effective noise canceling as perceived by humans?

And, again, that's not bashing FFT analysis. I genuinely do not understand how the video was interpreted that way, but it seems to be largely due to your conclusion that anyone's "bashing" FFT (which is not the case), and your insistence, then, on defending it to the death from an attack that hasn't occurred.

The conversation with Jacob in the video could open up a whole lot of interesting follow-up discussion about ANC, psychoacoustics, audio measurements, etc. I think we need to move past this idea that FFT has to be defended here, as it does not.

Fair enough @jude - I think we're disagreeing only in terminology and/or verbiage. I'm a total believer in Fletcher-Munson, but we can account even for the effect you mention, within the FFT that I'm no longer defending, with a simple weighting curve. I'll readily admit I don't know for sure what role psychoacoustics plays, but consider that every piece of hardware we know about on this planet (headphones, microphones, our ears) rolls off at the frequency extremes. Some part of the equal loudness story must for sure be hardware-related, rather than brain-related/psychological effects. Either way, I completely agree with the need to account for it. I've long suspected this as a source of disagreement amongst headphone owners who might be listening to the exact same headphones, but at different volumes.

 

[jude]

Fair enough @jude - I think we're disagreeing only in terminology and/or verbiage. I'm a total believer in Fletcher-Munson, but we can account even for the effect you mention, within the FFT that I'm no longer defending, with a simple weighting curve....

Like the A-weighting or C-weighting curves? I believe the A-weighting curve is an approximation of the inverse of the Fletcher-Munson 40-phon curve, which, again, is a psychoacoustic measure; and C-weighting similarly based on the 100-phon one.

I do look forward to seeing how this conversation develops, especially with the presence of several guys in this thread I personally know to be far more knowledgable than me on the topics at hand.

 

[sidecross]

A known problem of sharp minds is a tendency to cut each other up; It often does the opposite of sharpening.

Anyone who has worked with cutlery would know you sharpen an edge with a stone and hone with steel.

 

[arnaud]

A known problem of sharp minds is a tendency to cut each other up; It often does the opposite of sharpening.

Anyone who has worked with cutlery would know you sharpen an edge an edge with a stone and hone with steel.

This is such a funny analogy but I suppose there’s some truth to it .

I find the conversation in the thread fruitful though as it’s helping to clear some misinterpretations or help understand in more detail the measurement process for those (like myself) unfamiliar.

I do look forward to seeing how this conversation develops, especially with the presence of several guys in this thread I personally know to be far more knowledgable than me on the topics at hand.

Every time I read some updates on your endeavors with headphone measurements, I realize how much you’ve learned and, in some instance like here, how much I ignored or incorrectly assumed . I look forward to your continued partnership with the various measurement system manufacturers as it seems bound to lead to better practices on the ground by headphone manufacturers.

cheers,
arnaud

 

[castleofargh]

A known problem of sharp minds is a tendency to cut each other up; It often does the opposite of sharpening.

Anyone who has worked with cutlery would know you sharpen an edge an edge with a stone and hone with steel.

I bet 3 boxes of Pringles that @csglinux would be absolutely thrilled to have all sort of discussions with @pfzar and @Mr.Jacob. This one time it just happened to start on the wrong foot because Fourier is apparently a very touchy topic for him. I can understand, I also become quite sensitive when someone doesn't praise a member of my favorite boys band( it's ok Marie you can stay).

Last time I heard someone say that Plank wasn't fun at parties I got very mad.

Fair enough @jude - I think we're disagreeing only in terminology and/or verbiage. I'm a total believer in Fletcher-Munson, but we can account even for the effect you mention, within the FFT that I'm no longer defending, with a simple weighting curve. I'll readily admit I don't know for sure what role psychoacoustics plays, but consider that every piece of hardware we know about on this planet (headphones, microphones, our ears) rolls off at the frequency extremes. Some part of the equal loudness story must for sure be hardware-related, rather than brain-related/psychological effects. Either way, I completely agree with the need to account for it. I've long suspected this as a source of disagreement amongst headphone owners who might be listening to the exact same headphones, but at different volumes.

Even if we strictly limit the effort on statistical trends, we still need to try and factor music or a conversation on the phone going on at the same time. So now we need to count the average masking effect of the headphone's frequency response. We need to factor intelligibility of course(so even more insistence on the midrange than what the equal loudness contour would suggest IMO). But there is also going to be some concept of subjective annoyance of a given sound, which might not always correlate with the total energy or average loudness of it. Like how I can sleep just fine with the washing machine running in the other room, I'd even argue that it helps me falling asleep by masking other more punctual noises. but will go mad hearing the 3 little beeps signaling the end and repeating a few times every 30 seconds. The machine while running, has louder overall noise, even at the beeping frequency! But it's that fairly regular wide spread noise that the brain can treat as harmless ambiance and strongly reduce mentally. The beeps aren't like that.
Another famous example of stuff we easily push aside but only at a mind level, would be reverb. Even though it's a rather complex and obviously correlated signal, the brain will make use of the portion that can help figure out a position, and we will soon feel like the reverb has gone down at a conscious subjective level after standing at the same place.
So a precise model of all that would need to go really deep down the psychoacoustic rabbit hole before hopefully coming out with an objective predictive method.

Then there can be more case specific stuff going on. I think the most obvious example would be the voice of our mother. I can remain asleep with the windows open and people chatting and doing whatever down the street. But if my mother comes around, her whispering some village gossip to the nice old KGB granny next door, will instantly wake me up and have me fully focused on her voice. My mum's voice wouldn't have much impact on you. We all understand that, no matter our level of analysis. And we'll encounter other highly personal phenomenons like that, even though they usually won't be as extreme. Like how a noisy clock in a room drives me crazy but most people won't even notice until I mention it(then they curse me for pointing it out^_^). Or how some people won't notice the various noises on their vinyl playback and will feel like it's the best sound ever, while I can't wait for it to stop because all I can focus on is the background noise and possible "wow".
One of the possibly difficult aspect to deal with might simply be listening level. In part because we don't all listen at the same levels, but also and perhaps more importantly, because some will consistently increase the volume as outside noises increase, while other(me! Me, me!!!!) will not because they'd rather not hear something than ruin their hears to cover loud stuff. I think we can all see which trend we adopt by checking how loud our music is in the car when we come out of the highway. From those increasing the sound and leaving it high until next morning when it wakes up the neighborhood, to those who will turn the volume down as soon as they can, to those who will just turn off the radio and go stop the car to pass a call. My guts tell me that going loud and staying loud might be the majority, but I have nothing to back that up. If true, it would logically force us to reconsider the average SNR conditions.

Getting back to a more universal approach of perceived noises, I think no metric can be considered complete until it factors WMDs such as crying babies. I need a pacifier scale on my NC gears! Although passive isolation might be better for that frequency range? IDK.

 

[johnn29]

Thanks again for the interesting question. I hope my response was on point. But I'm well aware I might have triggered a series of follow up questions... !

Thanks Mr Jacob, that answered it well. I'm more familiar with the magic of things like convolution due to the Impulcifer project I use to simulate a 7.1 loudspeaker setup over headphones (created by a Headfier, Jaakkopansen) so it made sense. Amazing test bed you guys have created. I'm sure Bose used something similar to create the mic array on the 700 range which can distinguish between someone sitting close to you creating background noise and noise far away.

Also since you identified ANC in the 1-4k range as being super important you should try the Air Pods Pro with memory foam tips. They dramatically increase ANC in that region. See here for a plot. The amazing thing is that Apple is actively cancelling that range - it's not by a simple passive seal improvement. By far the biggest attenuation of I've tested on my own head even against regular foam earplugs.

I saw in the video you had the job title of Telcom Account Manager - I clearly missed a career path! I've spent 12+ years in IT/Telco as an Account Manager on the sales side and never had little opportunity to express my technical interest. Only to hit a quota!

 

[arnaud]

@Mr.Jacob , thank you again for the background info on MPNS and the related ESTI standard!
I’ve had a cursory look and is quite interesting! In particular, this is the first time I come across low pass filters with time variant characteristics (to avoid dealing with random room reflections at higher frequencies I suppose as this how I interpret these processed impulse responses).

Now, in regards to the loudness metrics, again it was helpful you mentioned about Moore-Glasberg as I was stuck at the Zwicker model from when I last looked into this, e.g. 20 years ago ;-p. I’d have to read up more but can I understand it as measuring loudness continously over the duration of the background noise playback and then integrate it to come up with a single long term loudness metric? In which case, the transient performance of anc (how it adapts to changing background noise) affects the metric you compute?

cheers,
arnaud

 

[sidecross]

I bet 3 boxes of Pringles that @csglinux would be absolutely thrilled to have all sort of discussions with @pfzar and @Mr.Jacob. This one time it just happened to start on the wrong foot because Fourier is apparently a very touchy topic for him. I can understand, I also become quite sensitive when someone doesn't praise a member of my favorite boys band( it's ok Marie you can stay).

Last time I heard someone say that Plank wasn't fun at parties I got very mad.


Even if we strictly limit the effort on statistical trends, we still need to try and factor music or a conversation on the phone going on at the same time. So now we need to count the average masking effect of the headphone's frequency response. We need to factor intelligibility of course(so even more insistence on the midrange than what the equal loudness contour would suggest IMO). But there is also going to be some concept of subjective annoyance of a given sound, which might not always correlate with the total energy or average loudness of it. Like how I can sleep just fine with the washing machine running in the other room, I'd even argue that it helps me falling asleep by masking other more punctual noises. but will go mad hearing the 3 little beeps signaling the end and repeating a few times every 30 seconds. The machine while running, has louder overall noise, even at the beeping frequency! But it's that fairly regular wide spread noise that the brain can treat as harmless ambiance and strongly reduce mentally. The beeps aren't like that.
Another famous example of stuff we easily push aside but only at a mind level, would be reverb. Even though it's a rather complex and obviously correlated signal, the brain will make use of the portion that can help figure out a position, and we will soon feel like the reverb has gone down at a conscious subjective level after standing at the same place.
So a precise model of all that would need to go really deep down the psychoacoustic rabbit hole before hopefully coming out with an objective predictive method.

Then there can be more case specific stuff going on. I think the most obvious example would be the voice of our mother. I can remain asleep with the windows open and people chatting and doing whatever down the street. But if my mother comes around, her whispering some village gossip to the nice old KGB granny next door, will instantly wake me up and have me fully focused on her voice. My mum's voice wouldn't have much impact on you. We all understand that, no matter our level of analysis. And we'll encounter other highly personal phenomenons like that, even though they usually won't be as extreme. Like how a noisy clock in a room drives me crazy but most people won't even notice until I mention it(then they curse me for pointing it out^_^). Or how some people won't notice the various noises on their vinyl playback and will feel like it's the best sound ever, while I can't wait for it to stop because all I can focus on is the background noise and possible "wow".
One of the possibly difficult aspect to deal with might simply be listening level. In part because we don't all listen at the same levels, but also and perhaps more importantly, because some will consistently increase the volume as outside noises increase, while other(me! Me, me!!!!) will not because they'd rather not hear something than ruin their hears to cover loud stuff. I think we can all see which trend we adopt by checking how loud our music is in the car when we come out of the highway. From those increasing the sound and leaving it high until next morning when it wakes up the neighborhood, to those who will turn the volume down as soon as they can, to those who will just turn off the radio and go stop the car to pass a call. My guts tell me that going loud and staying loud might be the majority, but I have nothing to back that up. If true, it would logically force us to reconsider the average SNR conditions.

Getting back to a more universal approach of perceived noises, I think no metric can be considered complete until it factors WMDs such as crying babies. I need a pacifier scale on my NC gears! Although passive isolation might be better for that frequency range? IDK.

The picture of the Solvay Conference does tell a story; being smart does have limits.

 

[Mr.Jacob]

Like how I can sleep just fine with the washing machine running in the other room, I'd even argue that it helps me falling asleep by masking other more punctual noises. but will go mad hearing the 3 little beeps signaling the end and repeating a few times every 30 seconds.

In addition to the white noise masking effects of the washing machine, this is also a great example of how our hearing is so sensitive to the changes in level. As you point out, the beeps are very quiet, but they "come out of nowhere" and really trigger our sense of hearing. The droning on of the washing machine fades (to black) and we don't pay much attention to it - EVEN IF it is objective louder.

@castleofargh also mentions the ultimate noise WMD: baby crying. Coincidentally, baby cry acoustic energy is usually in the 1k-5kHz range centered around 3.5kHz. Same as the smoke detector alarms: that's where our hearing is most sensitive! And aside from being a challenge for ANC headphones, now we also get into the great psychological examples mentioned (like your own mothers voice vs. a strangers voice etc.). One thing is the sensitivity of the ear and the way it responds to sounds, another is how we process them. That's a different discipline altogether and not one I'm trained in, but it's definitely tied to our discussions here and can really get us deep down the rabbit hole.

Speaking of WMD noise, for @johnn29, @arnaud and those of you interested in the standardized background noises, ETSI does have an open online repo - and they have binaural recordings from a Kindergarten if you feel like listening to that...! It is (not coincidentally) the second loudest A-weighted background noise scenario at an average of ~80.5dB(A) (only behind airplane cabin ~81.2dB(A))!
Binaural repo: https://docbox.etsi.org/stq/Open/EG 202 396-1 Background noise database/Binaural_Signals
8mic repo: https://docbox.etsi.org/STQ/Open/TS 103 224 Background Noise Database

The files are all in .wav so no special application will be required to play them back. However, the standards will tell you what absolute levels they are scaled to. These are have been used in telecom testing (mobile phones etc.) for years to help tune the noise suppression in the uplink direction. And now we are seeing more and more companies adopt and apply them for ANC test work.

 

[sidecross]

Thank you Mr. Jacob for your time and insight on this topic.

 

[Mr.Jacob]

Now, in regards to the loudness metrics, again it was helpful you mentioned about Moore-Glasberg as I was stuck at the Zwicker model from when I last looked into this, e.g. 20 years ago ;-p. I’d have to read up more but can I understand it as measuring loudness continously over the duration of the background noise playback and then integrate it to come up with a single long term loudness metric? In which case, the transient performance of anc (how it adapts to changing background noise) affects the metric you compute?

Hey @arnaud, yep, Zwicker is an easy example to use, because most in the industry is familiar with it (even the graybeards who have been doing this for 30+ years ), and today it seems a rather simplistic metric.
The Brians (Prof. Moore and Dr. Glasberg) have been cracking away at this problem for decades now, and as recently as 2016 updated their model again to also look at certain binaural effects. They are doing some fantastic work in this area.
Your understanding of the model is correct. They are running multiple concurrent analyses, where the instantaneous loudness values are smoothed over time, and using that to update both short term and long term loudness values. So it works for "Time-Varying" sounds, although I'm not 100% sure on how those smoothing functions account for the adaptive nature of human hearing. (This article might be helpful: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5318944/)
The Relative Approach hearing model (that is used in 3QUEST for instance) does a similar averaging over time. The "Relative Approach" name is actually an indication of how the model views the human hearing system: sensitive relative to the previously heard sounds. Hence why constant sounds (like a dishwasher/AC/appliance humming along) doesn't disturb us much (i.e. no change in level), but little beeps/squeaks/rattles or other transients can be very annoying/noticeable (sudden changes in level) even if they are quieter in absolute terms.

And of course, as you touch on, when you get into using realistic background noises as test criteria (mentioned in the other responses), you will be exposing the ANC headphones to non-deterministic and time-varying sounds, and the ANC will of course be responding to that, so it's important that we consider those temporal effects and changes.

Thank you Mr. Jacob for your time and insight on this topic.

Thanks @sidecross. I appreciate the interest and the good debate.

Thanks Mr Jacob, that answered it well. I'm more familiar with the magic of things like convolution due to the Impulcifer project I use to simulate a 7.1 loudspeaker setup over headphones (created by a Headfier, Jaakkopansen) so it made sense. Amazing test bed you guys have created. I'm sure Bose used something similar to create the mic array on the 700 range which can distinguish between someone sitting close to you creating background noise and noise far away.

Also since you identified ANC in the 1-4k range as being super important you should try the Air Pods Pro with memory foam tips. They dramatically increase ANC in that region. See here for a plot. The amazing thing is that Apple is actively cancelling that range - it's not by a simple passive seal improvement. By far the biggest attenuation of I've tested on my own head even against regular foam earplugs.

I saw in the video you had the job title of Telcom Account Manager - I clearly missed a career path! I've spent 12+ years in IT/Telco as an Account Manager on the sales side and never had little opportunity to express my technical interest. Only to hit a quota!

So, we're actually hiring right now for an account manager position.....
Fair warning, you'll be working with me. PM me if you want to hear more.

I will definitely be looking at the APP, and hope to test them in detail once the world (and the lab) opens up again. I hope to learn more about what they did in the APP and see how things look from an analysis perspective.

Thanks again for the great questions!

 

[castleofargh]

The picture of the Solvay Conference does tell a story; being smart does have limits.

You mean in term of sex appeal?




@Mr.Jacob after reading the papers you linked, I've come to the conclusion that I will win a Golden Globe in acoustic research for my idea of just applying R128 loudness normalization to both the noise and the signal we want to hear(separately recorded with binaural mics), and then simply classifying the disturbance by the dB ratio that the normalization applied between them.
I win, you lose, I'm a genius. Or maybe not.

More seriously, I'm wondering if there are known ANC approaches focused on noise shaping(so obviously leading to a louder overall noise), instead of just trying to attenuate everything as much as possible? Or maybe given the tiny distance between the mics and output drivers, we just cannot afford lag from computational work? but that should be easier to handle nowadays I imagine.
One specific noise I'm thinking of would be some rather consistent modulations within the noise like say we have mainly a 80Hz hum, but it fluctuates in amplitude with say a 0.5s second periodicity(or similar stuff we often find in public transportation). That's more annoying than just a stable 80Hz hum and some form of noise shaping target could probably do well in mitigating such sounds by reducing the amplitude of that specific periodicity. and that without having to need complex processes trying to identify it. Seems like I'm talking to myself but information or even educated guesses are welcome.