Understanding Audio Frequency Response & Psychoacoustics 

What an excellent video, I've learnt so much and it makes pefect sense. Explained very well.

Ive learned more in this 20 minutes about FR measurements, than from all other videos/guides from the past years.. thanks again for the great video Amir!

Excellent as always Amir! Thanks for the deep dive again.

Good day Amir...avid fan all the way from the Philippines. Thanks so much for the reviews and the knowledge that you give...keep it up and stay safe.

Fantastic video! I always wondered why I struggled to hear EQ changes at different frequencies. I thought my hearing was just getting worse. Now I know! ERB! I think I'll use the "ERB factor" when my wife accuses me of not listening to her. "But honey! ERB is preventing me from differentiating your higher pitched voice from other background noise" Think it will work?

Super informative. I've been viewing these Frequency Response graphs for some time but just now starting to appreciate how to understand them! Thanks Amir!

Should be required viewing to get an audiophile's licence.

i'm new to the forum and a complete noob in this field but i have studied both electronics and biology a lifetime ago so i really appreciate the way that you always make the connection between both fields (Psychoacoustics=biology in my mind). That connection is often missing in many analyses out there, which can be very missleading. Thank You!

Should be retracted. 7:14 "...variations here (around 240Hz) is far more audible because hearing bandwidth, ERB is very narrow" "...whereas, when you get up here (around 15kHz), you can see these little wiggles, your ERB is big chunk over here...these little peaks and valleys are never heard" "the same peaks and valleys here (around 240Hz) are (can be) audible." First of all, Amirm is clearly saying the word "same peak" on the log axis, otherwise, it does not look the same at all. And well, ERB's 51Hz bandwidth on 240Hz corresponds to about 0.306 oct and 1640Hz bandwidth on 15kHz corresponds to about 0.158 oct, obviously former is greater on the log axis. So, if you see the "same peak" on 240Hz and 15kHz on the log axis, the former has less possibility of audibility. I'm very sorry to say, because I'm paying respects for his efforts, but this time, I should say Amirm is completely wrong.

great mini tutorial! as always, YOU ROCK Amir!

Correction filters need to be split into few groups. 20-100, 100-1k and 1k to 3k. Above 3k it matters less and like said here, our ability to hear them is not great... and measurements just start to have more and more random elements the higher we go, and our methods of fixing small things in high frequencies is also not great. 20-100Hz is its own problem and we need this area to be smooth. Peaks and valleys here will affect the actual notes, you can have A1 to be loud and E1 to barely be audible. 100-1k is maybe the best range to fix, the impact is great but also our filters can do a lot. 1k-3k is starting to get more difficult to fix narrow problems but our hearing is quite sensitive. One school of thought is that when you have made big changes, halve them all.. When tuning speakers with ears this works the best, you fix the curve as best as you can and then you halve the amplitude of the changes.

This is terrific Amir - so helpful. I've been 'into' audio for 40 years and each of your videos makes me realise how much BS I have absorbed when, had I known better, all the while genuinely useful science like this was out there.

Thanks for explaining psychoacoustics in a reasonably easy to understand way.

This was brilliant!

Thanks a lot for your work Amir, your channel is proving to be an invaluable resource for us that want to understand more about audio reproduction and perception

Another excellent teaching video.

I can listen to you all day. Your soothing voice and plethora of knowledge makes a deadly combination

I was getting tired of hearing it's good for the price everywhere. Regardless of price is it good or bad, plain and simple. ASR keeps the recommendation simple and straight to the point. I can respect that.

Thanks,I appreciate your work.

Another excellent one Amir! Keep them coming.

Thank you Amir for your this basic detailed explanation. Would it be possible in the future to provide a deep dive into the equalization that you mention to assist in correcting errors, describing the hardware and/ or software options that are available, as well as the location within the audio chain where they should be applied?

Hi Amir. Thanks for the effort that you put into sharing this science. I had a thought about the comparisons of SPL fluctuations (around 7:50 in your video) and a further consideration is that when you look at the fluctuations (in your JBL example) around 230 Hz they span about 50Hz but the fluctuations at 13000Hz span maybe 3000Hz, owing to the logarithmic x-axis. Perhaps the fluctuations "seen" to be about the same size are actually close to equally audible, or at least the fact that the frequency response is plotted on a log scale would tend to "balance things out again" in relation to ERB?

Thanks Amir, always with the best explanation of complex subjects

Very interesting. This explains why even very low THD at lower frequencies can be easily heard.

Awesome topic and video! Thanks!

BTW Equal Loudness Contour was the FIRST THING we were taught in Audio Engineering School and for good reason. (I am a graduate of SAE NYC Class of 2004.)

Great topic!

Hold on a second! @7:20 you discuss small changes at ~250Hz are perceptible but the same width (on the graph) wiggles at 15kHz are not perceptible (due to ERB), but AND THIS IS A BIG BUT the plot is a log scale so the wiggles at 250kHz are about 30Hz apart (peak to peak) and the wiggles at 15kHz are about 4kHz apart (so ~100 x as wide on a linear scale) From the wikipedia equation @6:00 the ERB at 250Hz is ~27Hz and the ERB at 15kHz is ~1.6kHz. So, yes ERB sensitivity decreases with Hz but not at the rate you imply. In fact from these rough eyeball calculation the wiggle at 15kHz will be perceived more clearly than the one at 10Hz. That is just maths. So, in fact the standard log scale freq. response curve does in fact give a reasonable (for comparison sake) approximation of perceived fluctuations across the audible range. Very interesting subject though. Thanks for posting.

This is a GREAT video. Thank you. I am an EE, and this really appeals to my desire to measure and understand the measurements. It also is great for all those that want to claim that we do not understand sound and cannot measure all the things that affect how sound is perceived. Great video. I learned watching this video, and feel like I have been looking for this explanation for a long time.

It's sad but psychoacoustics is used to sell SOOO much hi end audio stuff. And to argue with some people that something can't do what it's said to do and they may be hearing placebo effect WOW! like you just called their mother dirty names. My latest argument with some people was about "audiophile" wall outlets! Yup, the recepticle in the wall, Oh, the bass got better more defined, better detail in the mids, on & on. I ask how and get told just listen for yourself. Pure nonsense.

You can't talk about psycho acoustics without mentioning the huge effect Equal Loudness Contours have on our perception of frequency response. I watched most of the video and no mention of it at all. You are no expert, but I appreciate the effort. But try harder next time before doing a video. (I finished the video and no mention of Equal Loudness Contours at all.)

Listen to a pair of speakers or headphones at 85 dB SPL and then listen it them at 93 dB SPL and you will clearly hear the effect of Equal Loudness Contour. Suddenly the same set of speakers or headphones that sounded flat is now harsh. The frequency response didn't change. Our perception of it changed due to the increase in SPL.

Thank you so much for this video and sharing your knowledge in this format, Amir! It is a great pleasure to learn from you and your experience. Very best greetings from Latvia! ^_^

So what you are saying is the horizontal scale should be even more "extreme". Thx

Actually, this was really informative and explains why bass management is so important. Thank you, Amir.

Is it true that ERB does not concern the human hearings ability to differentiate between levels of loudness? Let's say i had two clean sinus tones one at 10kHz and one at 100Hz. Now we make both tones x dB louder oder softer. I assume i would hear the db change in both tones with the same intensity because ERB is about differentiation in pitch not level. Am i wrong?

If high fidelity is convincing recreation of sound heard live from a recording then there are engineering criteria that must be taken into account in the playback system. The first arriving sound must have a flat frequency response all the way back to the recording microphones. Wide uniform dispersion particularly at high frequencies is critical. The beaming directional tweeters universally used today are awful. Then there are the reflections that must be engineered to recreate the reflections of live music. There are two kinds of problems, the reflections that would be heard were the musicians in your room and the reflections that would be heard at a concert venue. In the first case that is much of what you hear, in the second case that is almost everything you hear. Not only are spectral changes of reflections compared to the first arriving sound critical but so are the time delay, direction of arrival, and loudness compared to the first arriving sound. These must be engineered in a way to take into account the variables of the acoustics of the listening room and the variables in the way the recording was made and the kind of environment the type of music is best suited for. Since you can't hear the direction of even one reflection, understanding how the brain determines direction is critical if you are going to defeat it in an engineered sound field. So you have to learn how to model, analyze, and engineer sound fields. The currently accepted explanation of detection of direction is dead wrong. This is proven by binaural recordings played through headphones. It meets all of the believed theoretical criteria but fails anyway so it's wrong. So while frequency response is critical and necessary it is not sufficient. The current technology fails miserably and compared to live music. Recordings played through it sound dead, flat, boring compared the real thing. IMO the work product of this industry is a total failure. Did I solve this problem? Yes, nearly 50 years ago. I admit it's a tough problem, beyond the ability of the people I've met or listened to what they've had to say. I've solved tougher problems but this one isn't nearly as simple as they would have you believe. How can you solve a problem you don't even understand?

The statement between the lines is that FR needs to be smoothed according to our hearing. Surely those smoothing methods have been developed or are in development. And indeed I found that REW offers three methods, one of them being ERB-based. That would sound like a good choice. But two other (psychoacoustic and variable smoothing ) sounds it would try to do the same. The documentation suggests to use variable (VAR) smoothing when this will be the basis for EQ. Have you (or someone else here) tried the different methods? Why are there three? For different hearing conditions? Or is one better than the other? REW documentation excerpt: Variable Smoothing Ctrl+Shift+X Psychoacoustic Smoothing Ctrl+Shift+Y ERB Smoothing Ctrl+Shift+Z Apply a smoothing filter to the current channel. Repeating the action removes the smoothing. Variable smoothing applies 1/48 octave below 100 Hz, 1/3 octave above 10 kHz and varies between 1/48 and 1/3 octave from 100 Hz to 10 kHz, reaching 1/6 octave at 1 kHz. Variable smoothing is recommended for responses that are to be equalised. Psychoacoustic smoothing uses 1/3 octave below 100Hz, 1/6 octave above 1 kHz and varies from 1/3 octave to 1/6 octave between 100 Hz and 1 kHz. It also applies more weighting to peaks by using a cubic mean (cube root of the average of the cubed values) to produce a plot that more closely corresponds to the perceived frequency response. ERB smoothing uses a variable smoothing bandwidth that corresponds to the ear's Equivalent Rectangular Bandwidth, which is (107.77f + 24.673) Hz, where f is in kHz. At low frequencies this gives heavy smoothing, about 1 octave at 50Hz, 1/2 octave at 100 Hz, 1/3 octave at 200 Hz then levelling out to approximately 1/6 octave above 1 kHz.

WoWWWWW Amir when you talk you also say something! If you tell me something I learn something too, When you're done talking I'm smarter than before. If you're going to wear that blue-red suit with that cape and you put a letter A on your chest.. no one will ever look at Superman again. you are the only human on earth that i will pay to be friends with. very helpful this video when you are in the process of getting your speakers in place.

So basically the ear can only hear limited "colors" rather than working like a spectrometer

So if you can correct all speakers this way, why do some speakers sound better than others? Is there even a point in buying any kind of expensive gear?

the convoluted shape of the pinna (outer ear shape) influences the perceived sound in ways an audio cable is never be able to (or an amp for that matter). pun on a pair of over the ear phones, play a high fq sine tone and slightly reposition / move the headphones over one ear. you'll be able to attenuate the sound to such a degree that you might think you're having a defective headphone cable..

Thank you! The curtain is being pulled back for me....

Enjoy what you like and never doubt that you like it l

Brilliant presentation ---- thank you so much - you the Man

AUDIO GEAR SHILLS HEADS ARE BOILING, WITH EVERY VIDEO YOU RELEASE :D

Room correction algorithms can use thousands of filters and sound excellent.

Thanks for great video ! I was thinking applying AutoEQ convolution filter was best for my headphones. Do you then suggest to only use Parameter EQ with 5 bands that is also generated there ?

Thank you a lot Amir! Would you suggest then to apply "psychoacoustic smoothing" to the graphs?

Everybody interested in audio should be subscribed to your channel

This is another great article Amir, thanks. A question; since we have an ERB function, would it not make more sense to display the SPL, instead of on a logarithmic scale, on the ERB scale. We could then filter based on the bucket size and end up with a graph which represents the "average human" perceived response of the speaker. Has this been done? Would it make any sense?

Thanks, very informative!

Very interesting. I'm using Dirac on MiniDSP SHD Studio for 2 years now and every few months I'm changing target curves. Bud I did not know about looking at the curve differently between highs and lows. This is very helpfull. Still I would be very interested in a lesson :-) on adapting Dirac (or other) target curves.

Amir, well done. I learned a few things on this one; gonna have to watch a couple of times for it to sink in completely.

Great stuff, thanks for sharing!

Hello Amir, I have a many-year background in digital signal processing and acoustics and have a love for audio passed on from dad. I really appreciate the added psychoacoustics smoothing background and practical room and speaker equalization information- in a very palatable way. I will review your site more often. Great job, sir in helping to set a higher bar for those hobbyists who want to be more scientifically informed.

Amir thanks for all the useful information. I just purchased the Elac Debut Reference DBR62 speakers from Crutchfield. I was looking for a more neutral speaker for my next speaker. Hope it can fit the bill.

Thank you so much for that great explainer on how to relate frequency response to our hearing filters. This really helps me understand what has an audible impact and what does not in frequency response. In a future video, could you relate subjective musical sounds to the frequency response curve so I could make a visual map of what frequency response levels affect which parts of the "music sound spectrum." E.G. "From 2khz-4khz provide musical information about echos and spatial sounds that generate the recorded room or hall acoustic signature, and 20hz-30hz provide information on the XX instruments and reverberations creating physical effects felt by your body..."

I wrote a mail to Amir that I am leaving and closing my account. I have very musical ears and not much into audio tech itself. ASR is for gear fetches not necessarily trying to find stuff that sound good. I was the only female and senior member of that site, but I left because it is useless and a waste of time. ASR is corrupted with Chifi money. What they do is irrelevant and does not regard what sounds good. Those macho men strive towards extreme low THD numbers that has no audio-able significance. This is a site is not for music lovers, it is a site of angry middle aged and older men who do not like other opinions especial from a women. It is a sect without ears.

Hi Amir, I love my sound and Hifi and Im also Autistic. My hearing seems to be pretty sensitive. Is there any way of telling if I can hear in a different way to other people? Thanks

Thank you for this video. I have seen way too many DIY folks and software solutions apply a very large number of corrections to speaker frequency response with negative consequences. I believe this results in problems like you express along with the fact that these measurements can change dramatically with minor changes in microphone positioning. It is important to deal with problems that apply to the entire listening area and not a single microphone position. I also think large positive adjustments in equalization can result in increased distortion if the speaker struggles to handle the increase in power. I would suggest the use of wideband pink noise for observer evaluation of frequency response along with a wide selection of recorded sounds. Wideband pink noise will excite all frequencies over a short time frame while recordings are going to be more limited in content. This of course assumes the observer has some experience with what wideband pink noise should sound like. The only way to get that experience is to start using it on a regular basis. I have found that very narrow band errors of large positive magnitude at high frequencies can be very audible. I had a surround speaker in one application that had a very narrow band resonance at around 10 kHz and a width of no more than 100 Hz that was obvious as soon as I played something that excited that frequency. It was a sound that wanted to drill a hole in my head. Luckily I was able to remove it with a very narrow filter of about -30 dB magnitude. With that filter I was not able to hear any problem with the speaker. I was also not able to tell that the frequency was essentially not present in that speaker. I find large positive errors in frequency response of narrow bandwidth to be much more objectionable than negative errors. I think this is because most observers do not know what they are missing versus what is obviously exaggerated. An example of this is a friend of mine was at my home watching Hamilton. She is a professional opera singer and had memorized the film. She was astounded at the level of detail she heard in the orchestration in my home theater that had escaped her in other venues. She now plans to fly to my home to rehearse any new operas she performs if she can get her hands on a good recording. Highly resolving and reasonably accurate sound reproduction is unfortunately uncommon.

Amir, do you have information on the ERB vs. frequency from 20Hz to 20kHz? If so I would think that an algorithm could be made to produce a frequency response graph that shows the perceived frequency response vs. the measured frequency response.

Wow, apparantly I figured that ERB thing out before by accident, when I was doing exactly what you said about taking a random filter and applying it over different frequencies. I noticed I couldn't really hear a difference at higher frequencies and it makes so much sense now!

Thank you VERY much indeed! More of those! A question: Knowing the formula to calculate the ERB isn‘t there a way to clean up the measured frequence responce applying the ERB to the graph? Greetings from Germany

Thank you for helping to explain this topic. Would you mind creating some videos on the tuning of these zones? It seems like some rules of thumb (This Q for this range) would be helpful... or at least explaining why it’s not helpful would be... helpful? Thanks again!

Hello Amir. So, would the ERB smoothing be the best option in REW for measuring speakers frequency response (average room response) for a review? I wanted to use var smoothing according to other guide I've watched earlier but now I am not so sure - still a newbie in this. I'd be grateful for clarifying.

Thanks for this. What I don't understand is why is 1/3 octave too low resolution at low frequency? If at 20 Hz the bandwidth = 20 Hz, that is a full octave. So 1/3 octave = ~ 7 Hz would be 3x oversampled. Intuitively the widening of the auditory bands makes sense, as the musical scale is an geometric function. However, the low frequency explanation does not make sense to me. Sorry! If you can clarify that, I would very much appreciate it. I also find it fascinating that perception of volume levels is very coarse, while tone recognition is very tight. I hear to about 3-4 cents whether note is in/out-of tune, so that is about 1/200 of an octave.

What an excellent tutorial ! Great job!

Nicely done and perfect timing. Newbie question - As I'm looking to tweak things with my Sundara's I read in one of your charts "you need more current than voltage to drive the Sundara's" is that amps ? As someone who never listens at high volume what is considered more current and what is overkill (what am I looking for on the specs of my dac/amp) ? thanks in advance

Excellent presentation. With so much talk on smoothing and psychoacoustics, could you please comment on using the psychoacoustic smoothing filter in REW? Instead of guessing what may or may not be audible, doesn’t this filter automatically display what you’re actually hearing without guessing?

you forgot to think about one thing here, 3khz bank bandwith for example, dont mean its a smoothing filter in every case. it only filters when multiple sound are present there. a pluck of a string at 15khz still keep its volume, its when you add many other things in the same bandwidth it gets averaged out.

Amir, it seems from your talk that our hearing system's bandwidth is reasonably well known and you even touched on the possibility of an adaptive filter that would average wider and wider ranges as it moves along the frequency curve. This would be a great way of comparing the FRs of different speakers. What are the challenges in implementing this?

Hi Amir, love your stuff. Would appreciate a video explaining room modes, as you mentioned in this video. Thanks!

Excellent demonstration of psychoacoustics, Amir. Learned a lot. Thanks.

Amir, how can we implement room correction in between our source and DAC? What products are good for this? I’ve seen some stuff from Dirac but not sure how it works. Lots of us stream tidal or Qobuz. Is there something that we can use to smooth some of the low freq response at least ? Thanks !

One question, we all know with age that there is highi frequency hearing loss. But with age, does the width of the ERB change as well? In other words, as you age, do you start to average larger bands of frequency together? Thanks in advance!

Great video! !

good morning!!!

Always been curious about this...how do we get to things like imaging and stage width / height from a quantifiable measurement?

Excellent info. I remember the EQ craze in the late 70's and early 80's and the frustration customers had trying to improve the frequency response of their systems. Graphic EQ's were too blunt an instrument with the specific selected frequencies not necessarily in the correct range,due to what you just explained about broad range. There were quality Parmetric Equalizers from SAE , Soundcraftsmen and others but few really understood how to use them at the time or realized the changes would be subtle if properly implemented.

Great having reviews with measurements. It helps a lot to identify a badly designed piece of equipment. As a scientist my self I can tell that this is everything but science. Let not mistake what science is.

Thank you for this Amir, this has helped me understand much more now and my nerd meter is pegging pretty hard

Thanks 👍

Thanks a lot Amir. Very useful info.

Thanks for shining a needed light into the darkness.

Thx

Excellent, you've taught us so much. Thanks.

Very useful, thank you!

thank you for the public service! :D

♥

Very nicely explained, thank you!

Excellent! Learned something new!

Hi, Amir, great video! In your speaker reviews, you assess the effect of the applied EQ only by listening. Have you ever tried to repeat the Klippel measurement with EQ? I'm not sure if it's technically possible (I know you use Roon EQ so you'd have to insert another parametric EQ device into the signal path - if the Klippel system allows it).

Amir. Excellent. Could you please recommend one or two good reference books on psychoacoustics for those who would like to learn more on the subject. Also the discussion on width of the deviation. The graph is logarithmic and as such visually it accentuates width as a functional frequency. Would it not be better to adopt a horizontal scale based on ERB? Always enjoy your going back to school episodes.

I was hoping this would contain explanation of what headphones measurements imply what sound aspects. Qualities like detail, bass impact, airiness, veil/transparency, soundstage/imaging...

Thank you very much! Great lecture

Nice vid. I agree with eq filters, that less is better. Especially filters that boost. I've noticed distortion when boosting so I mainly try to attenuate.

Thx very much, this was eye opening. One question though - as in FR graphs the X axis is logarithmic, the graphs are already displayed in an ERB-similar manner. I.e., a certain dip at 100hz is roughly 1/10 the size, in hz, than a same drawn dip at 1000hz, as the X axis is much condensed at higher frequencies. So two visually similar dips should have same ERB effect?

Amir, is it possible to develop a method/standard of evaluating the quality of a speaker's frequency response, based on principles of ERB, Fletcher Munsen curves etc.?

Thank you so much! Very informative. No measurements or experience with ProAc Speakers? I have a pair of DT8’s and they measure pretty well in REW in room. Very very curious to someday soon try a pair of Revel PerformaBE to see what all the hype is about in very well measured speakers. Thank you once more Amir!