This is a huge topic. If you don't mind, I'd first like to get to grips with mikets42's method. Fancy measurements were a big part of my previous life.
But as an introduction, you might like to look at https://www.aes.org/e-lib/browse.cfm?elib=3798
The 'measurement' that makes a speaker sound like a speaker and not the real thing is my Room Interface Profile in that paper. After more than 20 yrs, I still don't know how to measure it though I have ideas. I'm hoping mikets42's method might give me more clues.
The other important factor is DBLTs Double Blind Listening Tests. If I was a guru in my previous life, it was a DBLT guru .. having used them to design speakers, mikes & electronics for some 2 decades
Yes. The technology is called Ambisonics and is a surround sound technique. I developed the Soundfield Microphone for it. You might want to look up papers by Benjamin, Lee & Heller for reports on the design of Playback systems including fancy use of DSP.
Hi Bluesystems,
I'm really sorry, but your idea that we need to take psychoacoustics into account is just a playfield of confusion for some high end manufacturers to make money with inferior products. After nearly 50 years in the field, I have seen it all.
Attempting to address an individual is a moving target, not even one person is a constant. What is a constant is the original sound. That can be measured against. I have said this many times, if we can reproduce that sound exactly, we have accomplished our goal completely. However a person hears that sound would be the same, so their hearing drops out of the equation. What changes is their mood, surroundings etc ... So you would have to make all that the same in order for them to hear anything the same way twice. Does this sound at all like a reasonable approach to you?
If someone wants to put an effects unit on their listening experience, okay. Fine for them. However, this distorts the sound and is only suitable for that person at that time.
Distortion. Difficult to pick out some times. I was used to LPs and tape machines, and FM broadcast. High distortion, compressed material. CDs opened that up and they were different. Obviously early reproduction wasn't good, but later generation machines can reproduce the mix with minimal distortion. Well, not unless you have a tube type D/A section! Now we have very high resolution files at high bit rates. Better than we have ever had it.
Recently I added a DAC to a streamer that had high distortion. It sounded fine (sort of) before, the spec being 0.5% THD (terrible). The DAC dropped distortion a lot, and while not glaringly obvious, the sound quality did improve immensely. Hmmm, distortion is absolutely important. Later I'll use a higher resolution streamer with that DAC, it should improve things a tiny bit.
My entire career has been centred on restoring and improving audio performance when i wasn't in the cal lab. The general public does recognise improved performance even when you don't tell them you did something. Measured performance these days does correlate to what people hear, it isn't the 1970's anymore. Products that suffer high distortion are also poorly designed (which is why they have high distortion). Most often "Designed by ear", a recipe for an inferior product without exception.
Also note that when most people post, they don't just run on and on. They use paragraphs, This would make your posts far more readable.
I'm really sorry, but your idea that we need to take psychoacoustics into account is just a playfield of confusion for some high end manufacturers to make money with inferior products. After nearly 50 years in the field, I have seen it all.
Attempting to address an individual is a moving target, not even one person is a constant. What is a constant is the original sound. That can be measured against. I have said this many times, if we can reproduce that sound exactly, we have accomplished our goal completely. However a person hears that sound would be the same, so their hearing drops out of the equation. What changes is their mood, surroundings etc ... So you would have to make all that the same in order for them to hear anything the same way twice. Does this sound at all like a reasonable approach to you?
If someone wants to put an effects unit on their listening experience, okay. Fine for them. However, this distorts the sound and is only suitable for that person at that time.
Distortion. Difficult to pick out some times. I was used to LPs and tape machines, and FM broadcast. High distortion, compressed material. CDs opened that up and they were different. Obviously early reproduction wasn't good, but later generation machines can reproduce the mix with minimal distortion. Well, not unless you have a tube type D/A section! Now we have very high resolution files at high bit rates. Better than we have ever had it.
Recently I added a DAC to a streamer that had high distortion. It sounded fine (sort of) before, the spec being 0.5% THD (terrible). The DAC dropped distortion a lot, and while not glaringly obvious, the sound quality did improve immensely. Hmmm, distortion is absolutely important. Later I'll use a higher resolution streamer with that DAC, it should improve things a tiny bit.
My entire career has been centred on restoring and improving audio performance when i wasn't in the cal lab. The general public does recognise improved performance even when you don't tell them you did something. Measured performance these days does correlate to what people hear, it isn't the 1970's anymore. Products that suffer high distortion are also poorly designed (which is why they have high distortion). Most often "Designed by ear", a recipe for an inferior product without exception.
Also note that when most people post, they don't just run on and on. They use paragraphs, This would make your posts far more readable.
This proves your attitude toward the matter. You hear a sound, you listen to the music.
One thing that you take for sure is the content of the disc, the original sound as one may call. And how can you attest that it sounds exact to the original sound ( I guess we're talking about music where there's something goin' on) ...DBT or something else?
Indeed sound is something that happens after hearing, if you follow...
It is based on FSAF (Fast Subband Adaptive Filtering) which is a 21st-century version of Welch tfe(). You can transform a full-band RIR into sub-band, back and forth, with NPR for any prescribed precision (-120dB reconstruction error is trivial). It takes much less MIPS re full band (which requires 2 hours to process 1 sec of recordings). By splitting the space, it can lower Fisher's matrix eigenvalue spread from 10^15 down to 10^4. I use single-model one-pass ReLS and full-band final filtering to avoid matching too well - otherwise, I would miss the dynamic instabilities of a driver.
So far, the best results (closest to the music) I achieved by using a modulated noise as excitation and examining the residual.
1. AWGN
2. shape it frequency-wise to approximate rock music (Queen's Dust in my example)
3. Modulate the amplitude
0-3 sec - nominal (I use -25 dBFS)
3-4 sec - nominal +10dB
4-8 sec - nominal -10dB
8-9 sec - nominal +5dB
9-12 sec - nominal -5dB
12-15 sec - back to nominal
I would say that the number of false positives (spk shows good numbers while it is not on real music) and false negatives (spk shows bad numbers while it is not) is much lower than for sine sweep, where correct answers and false positive/negatives are uniformly distributed, I have 2 | 3 dozens of speakers, which is not awfully representative.
Thank you, that's very nice work. I loaded a few publications and am reading them through.
Once upon a time, in the predigital era, I was designing SAW filters and had an opportunity to learn about multi-modal propagation, isotropic and anisotropic diffraction, Fresnel, Fraunhofer, etc. It also has acoustic waves (Relay) but is somewhat different from room audio.
There is a lot of hard science in psychoacoustics. This is real science and far far from the snake oil of some high end manufactures.
From Canada there is Floyd E Toole first from NRC and then the Harman group who I met and worked with on loudspeaker design for a number of years. With his research it was much easier to create loudspeakers that most listeners agreed sounded good. When you know what actually stimulates a positive listener experience it become much easier to create products that listeners find enjoyable. I learned a lot from Floyd that improved my art of audio design.
Psychoacoustics helps you avoid the common design trap of assuming that your current favored technical issue of "A" is the key to improvements in sound quality. Then technical issue "A" consumes most of your limited design time and as a result other more important factors to the listener's experience are not optimized as they could have been. The result is a prodcut that creates a listening experance that is wanting.
Psychoacoustics helps you know what you must do to make important improvements and how not to spend your time on what results in little to no meaningful improvement.
This relates back to the original question of this thread "Do measurements of drivers really matter for sound".
Psychoacoustics helps you answer this question. And the answer is "it depends" as I tried to expand on in my first post.
I think a large reason that voltage drive is most economical is because the semiductance provides a good inverse of the natural response rise that a cone has acoustically. I wonder if a lot of driver design is more about matching the cone and semiductance to create the most usable response than about any other consideration. And then I wonder what if you just screw the impedance curve and leave those considerations to the crossover/amp/DSP?
Even drivers with shorting rings often have a relatively flat response, so they must have chosen a cone with very different acoustic response, or maybe there is an aluminum coil former shorting the voicecoil.
Even drivers with shorting rings often have a relatively flat response, so they must have chosen a cone with very different acoustic response, or maybe there is an aluminum coil former shorting the voicecoil.
loudspeakers-for-aec-measurement-and-linearization
I was very glad to have access to Aaron Heller's copy when I presented https://www.aes.org/e-lib/browse.cfm?elib=14786 Using da right buzzwords goes a long way when U R pre10ding 2 B a guru
If I could ask for a few more clarifications ...
NPR: Near Perfect Reconstruction?
ReLS: ???
==================
Please correct my naive summary of your paper for people with only a single 'out of date' brain cell like me ...
Thanks for this Michael. I take it da buzzwords are from da MATLAB DSP toolbox.
I was very glad to have access to Aaron Heller's copy when I presented https://www.aes.org/e-lib/browse.cfm?elib=14786
Using da right buzzwords goes a long way when U R pre10ding 2 B a guru If I could ask for a few more clarifications ...
NPR: Near Perfect Reconstruction?
ReLS: ???
==================
Please correct my naive summary of your paper for people with only a single 'out of date' brain cell like me ...
- It ISN'T about Acoustic Echo Cancellation ie not about making pseudo anechoic measurements but about speakers that might respond well to this
- shaped noise signal is ramped up in level and then down.
- Measurement mike & low distortion LDC mike both in front of the DUT. Measurement mike used to 'flatten' LDC on-axis response ie correct 'linear' distortions with simple DSP
- Signal from da LDC 'de-convolved' with the 'input signal' using fancy new supa fast FFT methods
- Residual is presented in 'time-frequency' display to show how LTI varies with level
- Residual can also be listened to
NPR - Yes.
ReLS - Start with https://www.researchgate.net/publication/330853444_A_Shift_in_Paradigm_for_System_Identification .
Naive summary - correct. Without LDC, only by a typical measurement mic like Behringer ECM8000 you won't get > 50dB of dynamic range, which is a joke and a source of dystopian arguments.
BTW, you can get a kind of pseudo-anechoic measurement while selecting the very top of the ridge of the RIR
... (room impulse response) spectrogram (wavelet-ish). It works for well-damped closed boxes just fine but it is somewhat tricky to separate a vented box and a room. It does not matter how you get RIR.
ReLS - Start with https://www.researchgate.net/publication/330853444_A_Shift_in_Paradigm_for_System_Identification .
Naive summary - correct. Without LDC, only by a typical measurement mic like Behringer ECM8000 you won't get > 50dB of dynamic range, which is a joke and a source of dystopian arguments.
BTW, you can get a kind of pseudo-anechoic measurement while selecting the very top of the ridge of the RIR
... (room impulse response) spectrogram (wavelet-ish). It works for well-damped closed boxes just fine but it is somewhat tricky to separate a vented box and a room. It does not matter how you get RIR.
Honestly I do not take care too much of measurements.
With my DIY mods I change the characteristics of the (paper) cone, put a Faraday ring onto the pole plate and make even a whizzer by hand.
Only measured once frequency response to check if this cheapo driver can be converted to a fullrange driver or not.
https://www.diyaudio.com/community/threads/how-to-make-a-whizzer-cone.398063/post-7510218
https://www.diyaudio.com/community/...loudspeaker-sandwich-cone.402917/post-7443476
https://www.diyaudio.com/community/threads/how-to-diy-add-a-faraday-ring-to-a-loudspeaker.409852/
Then the EQ on my Fiio X1 or smartphone can correct roughly the response to desire. Additionally a transconductance amp can help for distortion and frequency response.
With my DIY mods I change the characteristics of the (paper) cone, put a Faraday ring onto the pole plate and make even a whizzer by hand.
Only measured once frequency response to check if this cheapo driver can be converted to a fullrange driver or not.
https://www.diyaudio.com/community/threads/how-to-make-a-whizzer-cone.398063/post-7510218
https://www.diyaudio.com/community/...loudspeaker-sandwich-cone.402917/post-7443476
https://www.diyaudio.com/community/threads/how-to-diy-add-a-faraday-ring-to-a-loudspeaker.409852/
Then the EQ on my Fiio X1 or smartphone can correct roughly the response to desire. Additionally a transconductance amp can help for distortion and frequency response.
With all the mods you can DIY - also adding extra same size magnets giving a plus of maybe 1.5db efficiency I tell friends asking for advice for throwing out standard loudspeakers in cars and putting new ones in that the original ones are not bad if you tweak them.
But I understand that many people do not know how to change / modify original drivers or fear the investment of time to apply the changes.
But I understand that many people do not know how to change / modify original drivers or fear the investment of time to apply the changes.
I (still) fail to see how you separate early reflections (mainly from ghost sources by diffraction and reflections) from the primary source. Or how you integrate the sum of all virtual sound sources on such short distance from the speaker, given the breakup modes you imply in your measurement.
To model how we do the process in our heads (with the intention of making the measurements better relate to subjective responses), the bispectrum and cross-bispectrum will reveal a lot of worthwhile information - the latter needing two microphones. Pertinent to other contributions above, I imagine sound field (matrix) filtering would be of great benefit here too - for those lucky enough to possess such microphones.
It will depend on whether the eddys causing the semi-inductive behaviour are flowing in steel or copper pole pieces, and how "stiff" is the magnet. Old fashioned AlNiCo magnets outperform more modern ferrite ones in the latter respect, for example.
But is "semiductance" a word or one you invented? I have never seen it used before, but I like it!! Does it subsume semi-inductance and semi-resistance in one complex lump, because using the term semi-inductance is often a bit vague?
This is a very profound statement - and IMHO the most succinct summary in this thread of the problems of relating measurements to how something sounds. Whilst I have been bringing up the bispectrum repeatedly as a means to model how we identify sounds and label them as belonging to some source or other, modelling physical sensation is only half of the process. The other half required to model a final perception would (mathematically at least) entail modelling a listener's entire lifetime experiences (learning) to have a clue as to the meaning they impart to the sensation - if indeed they are ascribing meaning to something they actually sense! Overcoming the potential for distortions in this latter part of any analysis is a formidable obstacle. It also provides significant ammunition to those using measurements as tool to influence sales.
Late to the party... What did I miss?
I also have some old Eton 7-Hex speakers lying around, waiting for a multi-way project, which also use titanium formers. Curiously, they also have a similar 'brightness' to the tone, but I can't be sure it's not coincidental. These days I'd go for a nice full-range anyway.
A fundamental issue with trying to emulate a point source is that it's not necessarily any more accurate than messing with the phase response, whether with flexble paper cones or dipoles or DMLs and so on, which tend to project various phase-shifted variations of a recorded snapshot at different angles. This ties in with an artistic question of whether you want to listen to something that sounds like a wall with a "peephole" to the actual sound source.
A concrete example could be an orchestral string section. Move the microphone slightly and you get a recording that's superficially very different because none of the bits match, but in practical terms the strings basically sound the same.
Conversely, a trumpet tends to focus an attention-grabbing beam of sound directly to the ears, and that's the kind of sound that I think Accutons could excel in -- planting a small live band in your living room, rather than a nostalgic whiff of a philharmonia in a theatre.
The super-smooth radiation pattern of ceramic can sound impressive. I don't know if there's all that much difference between ceramic and a heavy, metal cone though. Eddy currents in the conductive aluminium might contribute their own distortion. Also, I don't quite understand why Accuton didn't go all-out with non-conductive coil formers instead of titanium. The titanium obviously has to be cut to reduce eddy currents, but it's probably not enough to eliminate them altogether. And they probably give it lots of overhang, too, beyond the actual coil limits, to reduce modulation as the former moves around relative to the magnet, so that adds mass.
I also have some old Eton 7-Hex speakers lying around, waiting for a multi-way project, which also use titanium formers. Curiously, they also have a similar 'brightness' to the tone, but I can't be sure it's not coincidental. These days I'd go for a nice full-range anyway.
A fundamental issue with trying to emulate a point source is that it's not necessarily any more accurate than messing with the phase response, whether with flexble paper cones or dipoles or DMLs and so on, which tend to project various phase-shifted variations of a recorded snapshot at different angles. This ties in with an artistic question of whether you want to listen to something that sounds like a wall with a "peephole" to the actual sound source.
A concrete example could be an orchestral string section. Move the microphone slightly and you get a recording that's superficially very different because none of the bits match, but in practical terms the strings basically sound the same.
Conversely, a trumpet tends to focus an attention-grabbing beam of sound directly to the ears, and that's the kind of sound that I think Accutons could excel in -- planting a small live band in your living room, rather than a nostalgic whiff of a philharmonia in a theatre.