There is one on how they developed the EQ for a car (#127 I think) and found that using EQ curves that matched the ERB smoothed data was found to be preferred by a group of listeners. They compared this to 1/3 octave or something else I forget. It's just another piece of supporting evidence - hardly conclusive.
(PS Wow! what a body of work. A lot of it quite excellent to boot!)
(PS Wow! what a body of work. A lot of it quite excellent to boot!)
Last edited:
Barry, ear operates on variation of air pressure across time. Sure, inner ear has mechanism that functions in appearance much like spectrum analyzer with limited band differentiation; but this is no excuse for crushing measurements with broad smoothing.
Subjective domain is only correlated with objective domain with a level of statistical significance. There is no proper translation.
Real sounds and reproduced sounds can be measured and compared to finest level of detail.
Many real sounds are reproducible in manner indistinguishable to the human ear from the real sound.
You've got it backwards.
All measurements of human hearing system need to correlate with physics of sound propagation.
A good starting point is the small volume of air directly outside the eardrum.
This is followed by the resonant behavior of the ear canal. This is a built in source of large ripple in hearing range for which directional cues are intensity based. Below this range timing cues dominate, and these are ultimately represented by highly correlated fluctuation of air pressure in vicinity of outer ear canal at each ear. All this is highly validated by capture of HRTF data and its use in synthesizing recordings that produce highly realistic human perceptions of auditory scenes.
Loudspeaker transducers and associated fixtures often lead to highly complex radiation patterns across the human auditory spectrum. These manifest as response ripples with a high spatial dependance.
Even loudspeakers with high levels of this type of spatial distortion, when listened to in highly symmetrical setups are capable of producing a very good perception of an auditory scene. This sweet spot of perception is often quite small.
Speakers with lower ripple and less spatial dependance of shifting in peak frequencies produce wider and deeper listening sweet spot leading to listener assessment of higher quality.
All this is verifiable by measurements with sufficient resolution through range of frequencies most important to source localization of human hearing, roughly 500Hz-6kHz.
A typical living room is indeed quite sufficient for capturing relevant data. Use of too small a data window, inappropriate windowing function, or overly aggressive smoothing function of display data produces useless results.
Measurements for polar data must include enough points within region accommodating a human head with room for head movement. For example, assume a head region 12 inches wide. Then at a listening distance of 96 inches the aperture is roughly 7 degrees of arc. Uniformity across this span becomes quality criteria requiring multiple measurements. Polar plots with even 5 degree intervals lacks sufficient detail.
At 8ft distance and typical 3ft listening height, floor bounce angle is roughly 37 degrees with a path length of 10ft, leaving a scant 2ms of reflection free data. Truncating data with rectangular window is very poor approach. Windowing function such as four term Blackman-Harris window is better; this window can be somewhat longer than 2ms because terms at 2ms will be quite small. Small rectangular window is crappy gating of data. Truncating window and padding for bigger transform is poor practice that does nothing for the incurred jump discontinuity at the cut points.
Measurements from closer increase reflection free interval and increase direct to reflected intensity ratio. However speaker size and driver locations become more influential.
Problem? Your SHOUTING is Alpha primate posting behavior. Lots of posting, little real educational content. Not to worry, most Alpha primates are really sweethearts with deep and serious concerns for the well being of the troop....
Pardon my ignorance but I still don't understand how Farina's description of approaches to "room" EQ and ERB-like smoothing of gated impulse response speaker data for assessment of speaker performance characteristics relate to each other
Pardon my ignorance but I still don't understand how Farina's description of approaches to "room" EQ and ERB-like smoothing of gated impulse response speaker data for assessment of speaker performance characteristics relate to each other
It shows that of the smoothing techniques that he tried the ERB one correlated best o subjective preference.
This does seem reasonable.
Of course, those who already know this would see raw data for what it really means. Those who don't may appreciate it being reduced to what it actually means..if it weren't they may judge it based on their biases, or not. It also depends on the genuine intentions of the person posting the data.
This level of discussion would probably be the same whether on a forum about politics, or a forum on how to manufacture hot dogs.
Of course, those who already know this would see raw data for what it really means. Those who don't may appreciate it being reduced to what it actually means..if it weren't they may judge it based on their biases, or not. It also depends on the genuine intentions of the person posting the data.
This level of discussion would probably be the same whether on a forum about politics, or a forum on how to manufacture hot dogs.
Just that it is NOT what it actually means. I listen to my speakers in a room and not in an anechoic space...
It needs to be understood that in my case I have to smooth the data because I show the frequency points in log spacing, which of course only makes sense because or pitch perception is logarithmic. Hence there has to be a conversion from linear spaced FFT data to log spacing. Some smoothing is require at HFs to reduce the data set down to the log spacing. It only makes sense to pick a "logical" smoothing algorithm and not just grab 1/3 octave because that's what everyone else does. Unsmoothed data is not an option in my polar map.
And also remember that aliasing of the frequency response plot will occur if there are more data points than there are pixels. Do you just let this happen willy-nilly. I hope not.
And also remember that aliasing of the frequency response plot will occur if there are more data points than there are pixels. Do you just let this happen willy-nilly. I hope not.
Since high frequencies are being singled out I'll say that I find measurements for the top half of the spectrum for a typical free standing speaker most useful for determining the performance of the speaker alone. In the room, interactions would (or should) be in the far field and deserving of separate treatment where necessary. I'm not saying that everyone intends to see it that way but it does seem logical to try to keep the high frequency performance sufficiently independent of the room.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.