JPC, DD, at it again w/LTD02 stirring the pot...

[Bossobass Dave]

I was alerted to the recent bump of a thread at AVS started by Desert Dome regarding the difference between an RTA (Real Time Analyzer) and SpectrumLab, an FFT (Fast Fourier Transform) analyzer. Also differences in a logarithmic sine sweep vs a linear sine sweep and between pink noise and white noise.

 

There's lots of explanation of the differences in the analysis and the input signals.

 

The bottom line in the discussion for me is that the contention is that the entire library of collected and organized SpectrumLab data grotesquely misrepresents actual content on Movies With Bass. The SL caps have also similarly been described in terms varying from misleading to useless as a guide to what low end content is on a film.

 

In bumping the thread John (LTD02) said (among other things):

 

i created some pink noise in audacity and then observed the spectrumlab output. it shows the characteristic "increasing red" toward the lower end of the bass (+3db/oct amplitude arising from the wider bin width (more power per bin)). i then went into audacity and created a 3db/oct compensation eq and recorded the output in spectrumlab. 200hz is the reference point. quite a difference.

 

 

JPC had posted:

 

By definition, pink noise produces equal power (volume) per bin that are proportionally wide. For example each bin is 1 octave wide and has equal amplitude (volume). Pink noise is white noise filtered with the 3 dB per octave slope (AKA 10 dB per decade).

 

 

 

JPC posted:

 

 

So, if filtered noise, in this case pink noise, is fed into SL, it will accurately be displayed as having been filtered with a 3dB/octave slope. Making this as simple to understand as possible; if each "bin" is one octave wide, that means that 3 Hz to 6 Hz is one bin and 20 Hz to 40 Hz is one bin. Each of those bins has equal level across the bin. So, in the case of 3-6 Hz, 3 Hz, 4 Hz, 5 Hz and 6 Hz frequencies have a total level that the 21 frequencies from 20-40 Hz have. IOW, each frequency in the 3-6 Hz octave has a higher level than each frequency in the 20-40 Hz octave.

 

This higher concentration of level-per-frequency will [accurately] display as higher intensity on the SL color scale from 3-6 Hz than it does from 20-40 Hz........... IF YOUR INPUT SIGNAL IS PINK NOISE.

 

Congratulations, fellas! You've let us know what will happen if you use pink noise as the input signal into SL.

 

BUT WAIT... John then curiously adds:

 

what would be useful is a 3db/oct filter that could be applied prior to viewing content in spectrumlab, thus preserving all of its usefulness while providing a more accurate representation of content.

 

 

Because the low end content in MWB is mixed at higher levels than the rest of the audio band, is somehow mistakenly thought to mean that the low end content in MWB is mixed with a 3dB/octave filter, like pink noise.

 

Visit any audio recording studio and sit in on the mixing process. Because humans do not hear in a linear fashion, the low end is ALWAYS mixed hotter than the rest of the spectrum. The humans who do the mixing bump the low end until it "sounds right".

 

With movie soundtracks, explosions, erupting volcanoes, earthquakes, high powered weapons fire at close range, etc., the low end is bumped at the mix desk much more so than with music recordings.

 

Does this mean that, as John asserts, we need to apply a filter to the content in order to view it properly in SpectrumLab?

 

Fuck no, it doesn't. Are you shitting me, or what?

 

We don't want to see the content in SL as humans hear it. We want to see the content in SL as an accurate spectrograph of the actual content. Otherwise, hold up your 1/3 octave RTA and look at that graph while you play the movie and skip these forums altogether.

[Bossobass Dave]

Then, DesertDome chimes in, taking bows for John's accolades and posts a SL cap of the EOT opening scene posted by James (MKTheater), which is a mic'd-at-his-seats version. Here is the posts:

 

 

This is the kinda stuff we have to contend with. The wrong and flawed data twisted to conform with the wrong application to our SL content caps.

 

This prompted Paul to do a more thorough examination of the EOT scene, using SL and his new, extremely accurate digital Oscilloscope.

 

With all due respect to MKT, who has not made any claims as to the relative accuracy of his MIC'd-At-THE-SEATS SL cap on a frequency-to-frequency basis, and to the accuracy of the freeware audacity or any of it's users, Paul's data is accurate to within microvolts.

 

In the next post, we'll examine the scene and the actual difference in voltage from the 25 Hz fundamental and the 10 Hz fundamental square waves and their associated odd-order distortion in toto and how that translates to dB levels and comparing that to the SL spectrograph and waveform renderings.

[Bossobass Dave]

Here is the actual data for the opening scene of EOT...

 

First, what the SpectrumLab spectrograph actually shows. Paul has recently been messing with color scale settings, but that is irrelevant to what SL will show in it's spectrograph and it's waveform graphs.

 

 

 

Here is the waveform from the digital 0scilloscope and the recorded difference in voltage between the 25 Hz fundamental and 10 Hz fundamental square waves. Also included are the oscilloscope captures of the actual 2 square waves. I put this into a gif to show the waveform differences between the much higher-rez O-scope vs the SL versions:

 

 

 

Of course, the SL color graph is not accurate enough to show dB differences to the degree the calculator O-scope combination show, but around 5dB difference was calculated from the measured voltage difference and the color scale accurately reflects that. Nor is the SL waveform graph able to capture transients nearly as accurately as the digital O-scope, but, again, it should be painfully obvious to any casual observer that the differences are not of any significant magnitude, let alone 12-15dB, as claimed.

 

IOW, F-A-R from the 15-20dB difference DD claims and F-A-R from requiring a filter in SL to "properly view the content".

 

It's actually the 15 Hz fundamental square wave that's encoded the loudest of the 5 and the 30 Hz square wave that's the lowest in level, but that is also irrelevant. What is relevant is that SL accurately shows the content on the disc. It is not tilted at 10dB/decade from DC-120 Hz, the bandwidth of interest.

 

In the following posts, we'll set the Raptor System III subwoofer as flat as possible, in-room, from 10 Hz to 30 Hz and film and measure the output differences of the 5 steps of the effect to show that what goes into the room as reproduced by the SW is exactly the same as what is encoded onto the disc and what the SL waveform and spectrograph graphs quite accurately depict.

 

We'll analyze this scene to microvolts and adjust and zoom the SL spectrograph to one color per dB if necessary to dispel this hypothesis that all of our hard work on MWB is so flawed as to be irrelevant to what we should expect on playback.

[Bossobass Dave]

LTD02 has no subwoofer and never has, so his subjective opinion regarding what the SL caps show vs how the soundtrack will be presented to our persons by the SW system is irrelevant.

 

JPC and DD have systems that eliminate 2/3 or more of the ULF and present the top 1/3 at below reference levels so, again, whether or not our caps give them a heads up as to what to expect from the listening experience is... irrelevant. In fact, even if you have a perfectly flat response to 3 Hz at reference level with 10dB of headroom, whether or not our data is valuable or how it may be of value to your expected personal listening pleasure has nothing to do with this subject.

 

We just want to show that our accumulated data for MWB, which consists of peak hold and average graphs as well as spectrographs and waveform graphs and a rating system that 75% based on measured data for dynamics, level and extension, is accurate to what is encoded on the disc.

 

We occasionally post excursion vids that should be enough proof that ULF is not depicted in SL as being higher level than it actually is. There should be no flat-earther's denials of the excursion required vs what the spectrograph shows. Nor will the delusion that a single SVS PB-13 Ultra can accurately reproduce any of these movies at reference level in any room using whatever flawed reasoning get a pass.

 

There is a single 25% vote for subjective preference that is biased by the listener and his system and that is not in question here.

[Bossobass Dave]

Next, we analyzed the 10 Hz and 25 Hz voltage of the signal input into the SW amplifiers.

 

1. Measure frequency response at the main seat to verify flat response at 10 Hz and 25 Hz.

 

2. Play 10 Hz & 25 Hz and record level at the main seat with SPL meter, 'Z' Weight, Slow.

 

3. Leaving mic at the main seat, record 10 Hz & 25 Hz sine tones through microphone measurement rig into SpectrumLab.

 

4. Measure voltage of signal being fed into the Raptor System III amplifier using digital O-scope.

 

 

Conclusion: SpectrumLab shows same level spectrograph of each of the 2 tones as they were input (and verified by SPL meter, O-scope voltage measurement and in-room frequency response of the Raptor System III. It does not show an exaggerated 10 Hz spectrograph vs the 25 Hz spectrograph, nor any perceivable difference in the waveform graph.

 

This test can be repeated using 3 Hz vs 100 Hz and will show the same results, not a 20dB exaggeration, as claimed.

[SME]

It sounds like they are just confusing themselves.  I don't know SL in particular all that well, but I am quite familiar with the analytical methods it uses.  There is no need for correction of the data as far as the content is concerned.  As your testing reveals, the spectrograms are completely consistent with the output voltages you are measuring.

 

What is needed is careful interpretation of that data in terms of what we actually hear.  The bottom line is that the spectrogram does not reveal exactly what we hear, and no 3 dB/octave "correction" will change that.  What we hear is much harder to characterize and may even vary significantly between individuals.  Many studies have suggested that our ear acts as a bank of filters whose width varies in frequency but not in a way that resembles the frequency weighting of pink noise.

 

A most notable example of this work is the Bark scale, which roughly predicts the propensity of tones to mask one another, depending on their separation in frequency.  A louder tone within a particular Bark bank is more likely to mask a softer tone in the same bank.  Note that relative loudness is described approximately by the equal loudness contours.  Together, the Bark scale and ELCs (which also don't resemble 3 dB/octave curves at all) are used along with temporal descriptions of our hearing to develop the psychoacoustic models used in lossy audio data compression codecs and algorithms.  In practice, these codecs (i.e., Dolby AC3, MP3, AAC, Vorbis, and later formats) have been very successful, which provides strong support for these models.

 

It is notable that the Bark scale suggests that the 1st of 24 critical bands covers the entire sub bass range (< 100 Hz).  I'm not sure if this is because Bark did not study these frequencies in sufficient detail, or if our ability to discriminate bass is really that poor.  I'm not aware of any good follow up studies on this point.  As I've said before, there is little motivation to study the hearing of sub bass frequencies when designing audio data compression codecs because very little data is required to represent content in these frequencies in the first place.  If our ability to discriminate bass is as poor as the Bark scale suggests, the implication is that flat response is even more essential for bass than for higher frequencies, both in the mixing and playback processes, to ensure all intended content is heard because masking occurs more readily down low.

 

The spectrograms do not account for loudness perception differences as indicated by ELCs nor do they account for the effects of masking, which can be quite complicated.  Which model should be used to convert from a raw spec to "what we hear" is a moving target as this research is still on-going.  Each new generation of lossy audio data compression technology introduces a more sophisticated psychoacoustic model of masking, which leads to a further increase in performance.

 

Another thing to consider with the spectrogram is the optional parameter that dictates how much resolution to present in frequency versus time.  This is dictated by the width of the Gaussian window in the short time Fourier transform (STFT).  A wider window will provide more frequency resolution at the expense of time resolution, and visa versa.  There is no single correct window setting to use.  Different settings present the same data in different ways.  Our hearing is very flexible and is able to both identify transients with high temporal accuracy and identify continuous tones with high frequency accuracy.  To really understand what's being heard, one should look at spectrograms using a variety of window widths.  The narrower windows better reveal how we are likely to hear transients, and the wider window better reveal how we are likely to hear continuous sounds.  I understand that posting specs with different window sizes is a bit impractical, and I gather the specs you post here rely on a single window width that represents a reasonable compromise.  I'm not sure if this window width was consciously chosen or if you all simply went with the SL default.  (If that's how this works.  As I said, I'm not familiar specifically with SL.)  This is fine, as long as we recognize the limits of this process.

[Bossobass Dave]

The window and scroll speed were chosen to reflect a higher degree of frequency resolution because the waveform graph on the right side of the spectrograph shows the time domain information in very good resolution, for those who prefer a tilt toward the time domain.

 

Yes, it's still a compromise, but is tilted toward frequency resolution because we are primarily interested in frequency content at what level in the 3-120 Hz bandwidth.

 

The color scale was chosen solely to closely reflect that used by the majority of early SL posters to avoid unnecessary confusion.

[tuxedocivic]

 

 

Visit any audio recording studio and sit in on the mixing process. Because humans do not hear in a linear fashion, the low end is ALWAYS mixed hotter than the rest of the spectrum. The humans who do the mixing bump the low end until it "sounds right".

 

With movie soundtracks, explosions, erupting volcanoes, earthquakes, high powered weapons fire at close range, etc., the low end is bumped at the mix desk much more so than with music recordings.

 

 

This is such an important point. I get this all the time with speaker design and house curves. People forget that bumping up the bass has already happened in the recording. I guess they either like it even hotter than the mixers do, or they think it's their requirement to make the bass sound right. The Fletcher Munson curves are a huge part of this. They should rarely enter the conversation, but for some reason people think they need to compensate for what Munson found. Newsflash, the mixer/musician already did it

 

Not exactly on topic, sorry, but I do appreciate what you're saying here. I think it comes down to this: People want to dismiss ULF and justify high pass filtering 2+ octave of bass cause ported = loud + cheap. So they look for ways to dismiss the evidence when they have none of their own.

[SME]

The window and scroll speed were chosen to reflect a higher degree of frequency resolution because the waveform graph on the right side of the spectrograph shows the time domain information in very good resolution, for those who prefer a tilt toward the time domain.

 

Yes, it's still a compromise, but is tilted toward frequency resolution because we are primarily interested in frequency content at what level in the 3-120 Hz bandwidth.

This sounds like a reasonable compromise as far as an overview of the content is concerned.  Not only do you need the wide windows to see ULF detail, but you are often interested in content for scenes that span 30 seconds or more.  Unless you are willing to zoom in more, the extra temporal details that appear when using a narrower window may not appear anyway.

 

That said, I often see specs that don't intuitively "look like" what I heard in the effect.  For example, I think the bass in the wormhole scene in "Interstellar" sounds a lot tighter and more structured than it appears in the spec you posted.  The spec makes it look like there's just a bunch of narrow band noise centered near 30 Hz with distortion products at higher harmonics.  What I heard and felt in that scene was more like being physically jostled by turbulent forces with many "bumps" being felt distinctly from one another.  That feature does not obviously appear either in the spec or in the temporal waveform shown to the right of the spec.  (I reckon the temporal waveform does not display with enough resolution to see this.)  I believe the transient content was primary to my perception.  The heavy emphasis near 30 and 60 Hz was only a secondary contribution as it probably contributed the impression of the space ship hull ringing after each bump.  The additional content appearing in the spec around 90 Hz and 117 Hz is more structured and offers a hint that there's more going on in this effect.  If I'm reading the color scale right, it looks like the content at 90 Hz is about 18 dB lower than the content near 30 Hz.  Taking into account loudness (ELCs) we can expect the content at 90 and 120 Hz, independently, to be of similar loudness to the 30 Hz content, so their contributions to what we hear (on a properly calibrated system) are about equally important as the 30 Hz part.  I wouldn't be surprised if there is content above 120 Hz that also contributes a lot to the sound effect.  On the other hand, rendering the data with a narrower time window (and probably zooming in a bit on the time axis) may make that temporal structure more obvious.

 

Edit:  I just realized that I was looking at the spec capture from a clipping Oppo.  The non-clipped spec does indeed show more structure in the time domain data shown to the right of the spec.  OTOH, I believe the pure time data still does a poor job of revealing the time domain structure that we actually perceive.

[LTD02]

hi dave,

 

the fft requires 3db/oct adjustment when the content *spans* the frequency range, as it does with typical movie effects (explosions, rumbles, etc.) or pink noise because as bin widths get progressively wider the lower the frequency, summing over wider bins captures more energy per bin (which results in a higher level and brighter color in spectrum lab).

 

when there is a *pure tone*, as is the case with some movie effects, the fft bin width doesn't matter because all the energy is in a single tone.  with pure tones, no adjustment is required because there is the level of the tone and then just a bunch of empty space in the rest of the bin.

 

i've confirmed both these effects.

 

so what to do?  with rumbles and such, 5hz content will appear 13db hot relative to 100hz content even when recorded at the same level (dd's point in the avs thread), but including a 3db/oct correction curve will cause 5hz sine waves to appear 13db lower than 100hz content even when recorded at the same level (your point).  to the extent that movie content contains a mixture of rumbles and pure tones, there would appear to be no right answer--only two "less than ideal" answers. 

[Bossobass Dave]

hi dave,

 

the fft requires 3db/oct adjustment when the content *spans* the frequency range, as it does with typical movie effects (explosions, rumbles, etc.) or pink noise because as bin widths get progressively wider the lower the frequency, summing over wider bins captures more energy per bin (which results in a higher level and brighter color in spectrum lab).

 

when there is a *pure tone*, as is the case with some movie effects, the fft bin width doesn't matter because all the energy is in a single tone.  with pure tones, no adjustment is required because there is the level of the tone and then just a bunch of empty space in the rest of the bin.

 

i've confirmed both these effects.

 

so what to do?  with rumbles and such, 5hz content will appear 13db hot relative to 100hz content even when recorded at the same level (dd's point in the avs thread), but including a 3db/oct correction curve will cause 5hz sine waves to appear 13db lower than 100hz content even when recorded at the same level (your point).  to the extent that movie content contains a mixture of rumbles and pure tones, there would appear to be no right answer--only two "less than ideal" answers. 

 

Hi John,

 

You're confused.

 

Bin width is not 1 octave, it's in the millihertz range. Content does not appear 13dB hot at 5 Hz compared to 100 Hz when the proper settings are used with the proper window.

 

In my example SL cap of a scene from B:LA, the pink blip at 93 Hz is not +13dB hotter than the pink blip at 5 Hz. If that were the case, the drivers would not have been at full excursion with the amplifier's current peak limiter lights flashing and neither would the peak hold graph have shown a corresponding peak and my 27' x 24' wood frame floor would not have rippled.

 

 

1) There ain't no PB-13 Ultra gonna play back this scene with no problems. There's a reason we use multi-15" long throw drivers, signal shaping boost and multi-kilowatt amplification.

 

2) There's a reason that MWB scenes are peaked at 30 Hz and it isn't because SL exaggerates any specific frequency. It's simply because humans mix the sound and their human hearing dictates that they bump the low end. Since none of their playback systems give much <30 Hz, they may filter the soundtrack at 30 Hz or just leave the <30 Hz content intact with no cut or boost.

 

SpecLab is not doing the boosting.

 

3) The work done here is a phenomenal and mammoth on-going undertaking for a few guys who do this as a hobby and offer it free, non-ad-polluted, non-censored and fun as hell. Telling people to basically ignore it because it's data is errant by 30 times or worse is bullshit. Telling people to just buy a ported ID sub to handle WOTW would be shitz 'n giggly if it weren't downright insulting to decades of debating, designing, building, testing and verifying what is required to play back a recording like the WOTW soundtrack at reference level in-room.

 

Otherwise, I couldn't (and generally don't) care less what your opinion is about low end reproduction.

[Shredhead]

I made a track of 20 sine waves all at the same level in the DAW from 5 to 100Hz.  I played it back and at the end I held out 5Hz and 10Hz to make sure that there was no change in the levels when the low tones were held by themselves without the others.  There seems to be no difference for the level SL shows for the fundamentals in the spectrogram and the peak/average graph.  The red average goes higher than the other tones for 5 and 10Hz because they are held for a longer time period. 

 

 

 

I don't see this frequency spread test being too much different from source content.  John, maybe your settings on SL can be tweaked for higher resolution so you aren't having the tilted response problem. 

[SME]

the fft requires 3db/oct adjustment when the content *spans* the frequency range, as it does with typical movie effects (explosions, rumbles, etc.) or pink noise because as bin widths get progressively wider the lower the frequency, summing over wider bins captures more energy per bin (which results in a higher level and brighter color in spectrum lab).

The bin widths do not get progressively wider as frequency decreases.  In the FFT or (as in the case of spectrograms) the STFT, the bin width is constant with frequency.

[Madaeel]

I don't get this either. I've never seen a graph where I thought the low end was exaggerated 3-10db. It always looks like what I heard and felt. Another example is Cloverfield when the monster steps down. Content out to 85hz and all the way down to 15hz is the same level except the slight blip at 27hz.

 

So either it was encoded at the same levels or they're saying the low end should be 10db lower. When you have subs that extend that low you do get used to how it should sound and feel. I see nothing wrong with this graph and if indeed they're right there's no way you would feel much of the content down there. There's plenty of examples like this too.

[Bossobass Dave]

I made a track of 20 sine waves all at the same level in the DAW from 5 to 100Hz.  I played it back and at the end I held out 5Hz and 10Hz to make sure that there was no change in the levels when the low tones were held by themselves without the others.  There seems to be no difference for the level SL shows for the fundamentals in the spectrogram and the peak/average graph.  The red average goes higher than the other tones for 5 and 10Hz because they are held for a longer time period. 

 

 

 

I don't see this frequency spread test being too much different from source content.  John, maybe your settings on SL can be tweaked for higher resolution so you aren't having the tilted response problem. 

 

Excellent post. ^^^

 

The bin widths do not get progressively wider as frequency decreases.  In the FFT or (as in the case of spectrograms) the STFT, the bin width is constant with frequency.

 

Yes, and Paul's experiment with the multi-tone audio file is perfect proof. My settings are approximately 1 quarter of a hertz (250 millihertz) FFT bin size. The density does not change as frequency decreases, as you've pointed out.

 

 

I don't get this either. I've never seen a graph where I thought the low end was exaggerated 3-10db. It always looks like what I heard and felt. Another example is Cloverfield when the monster steps down. Content out to 85hz and all the way down to 15hz is the same level except the slight blip at 27hz.

Cloverfieldsubway.jpg

 

So either it was encoded at the same levels or they're saying the low end should be 10db lower. When you have subs that extend that low you do get used to how it should sound and feel. I see nothing wrong with this graph and if indeed they're right there's no way you would feel much of the content down there. There's plenty of examples like this too.

 

Yes, we KNOW the low end isn't -13dB lower at 5 Hz than it is at 100 Hz, right? But, remember, you actually have to have a subwoofer and one that plays 5 Hz at reference PLUS 10dB for this to become obvious during playback of MWB.

[Bossobass Dave]

This is such an important point. I get this all the time with speaker design and house curves. People forget that bumping up the bass has already happened in the recording. I guess they either like it even hotter than the mixers do, or they think it's their requirement to make the bass sound right. The Fletcher Munson curves are a huge part of this. They should rarely enter the conversation, but for some reason people think they need to compensate for what Munson found. Newsflash, the mixer/musician already did it

 

Not exactly on topic, sorry, but I do appreciate what you're saying here. I think it comes down to this: People want to dismiss ULF and justify high pass filtering 2+ octave of bass cause ported = loud + cheap. So they look for ways to dismiss the evidence when they have none of their own.

 

Also a great post, Tux. I agree and It's given me a chuckle for 40 years, watching people run to my preamp to "Re-Mix" the mix with whatever available EQ tweaks were available in the rack.

 

How can you tell the difference between a good mix and a not-so-good mix? Have a flat playback response... period. After that, distort the playback response however it suits you, but always have the reference at the ready.

 

This is also in line with my post about the ever-changing FR caused by aggressive on-board limiters on commercial HT subs.

[Bossobass Dave]

Here's another view of the situation. Take Adam's Raptor System III 8 x 15" drivers and calculate excursion at 30 Hz at 110dB, look at the excursion requirement and then at 7.5 Hz at 110dB and look at the excursion requirement. Then, reduce the output at 7.5 Hz by -13dB and watch the drivers back off to 1/2" PtoP.

 

 

So, Adam, which excursion/output numbers look right to you during playback?

[Madaeel]

Haha I wish it was 7mm! I would do that scene on ST at 30db hot.  

[LTD02]

our hearing is log sensitive, so we use log signals (log sine sweeps, pink noise) to eq sound systems.  this way what appears on the screen and what we hear correlate.

 

log sine sweeps and pink noise have decreasing energy at 3db/oct.  as a result, spectrum lab, which presents data differently than the typical audio rta, presents such signals will have a 3db/oct downward slope.

 

as a result, in spectrum lab, when there is equal energy on the disc in the octave around 5hz, it will present 13db higher than the same amount of energy in the octave around 100hz (and another 10db higher than 1khz, and another 10db higher than 10khz, etc).

 

the tones that shredhead posted are showing equal energy per tone.  in spectrum lab, such content will appear to be at the same level.  we all agree on that, i hope.  of course the higher frequency tones will sound much louder, which is why we don't use this method to eq sound systems.  we all agree on that, i hope too.

 

the bottom line is that most folks who look at the spectrum lab captures are thinking that it is presenting data in the same way as the typical audio rta.  it is not.  the typical rta has the -3db/oct adjustment baked in, spectrum lab does not.  i'm not really sure that either could be called more or less "accurate", but one correlates with the way that we hear while the other does not.

[Bossobass Dave]

Sorry John, most people are viewing the objective scores, peak hold/average graphs and spectrographs to see what DC to 120 Hz content is on the soundtrack because those 4 metrics accurately convey the content with numbers, traces and pictures..

 

Log sine sweeps and pink noise are irrelevant, unless you want to view the content of such accurately in spectrograph form, which no one does.

 

We don't use Real Time Analyzers here, in case you, JPC and desertdome haven't noticed. It's the wrong tool for the job.

 

desertdome made the assertions that the SL caps we use here misrepresent what's on the soundtracks to the extent that the SVS PB-13 Ultra sub in 15 Hz tune could easily replay WOTW to 14 Hz (or whatever frequency he said) at reference level, that all of the peak hold graphs posted here need to be tilted to be viewed correctly and that single digit content is exaggerated 15-20dB in the spectrographs posted here vs what he claimed the actual level is.

 

You agreed with him and congratulated him on his data and even suggested applying a filter in the settings when making the SL caps so the spectrographs could be viewed properly.

 

I called bullshit.

 

Hopefully we all can agree that backpedaling is not the right tactic at this point.

[LTD02]

Sorry John, most people are viewing the objective scores, peak hold/average graphs and spectrographs to see what DC to 120 Hz content is on the soundtrack because those 4 metrics accurately convey the content with numbers, traces and pictures..

 

Log sine sweeps and pink noise are irrelevant, unless you want to view the content of such accurately in spectrograph form, which no one does.

 

We don't use Real Time Analyzers here, in case you, JPC and desertdome haven't noticed. It's the wrong tool for the job.

 

desertdome made the assertions that the SL caps we use here misrepresent what's on the soundtracks to the extent that the SVS PB-13 Ultra sub in 15 Hz tune could easily replay WOTW to 14 Hz (or whatever frequency he said) at reference level, that all of the peak hold graphs posted here need to be tilted to be viewed correctly and that single digit content is exaggerated 15-20dB in the spectrographs posted here vs what he claimed the actual level is.

 

You agreed with him and congratulated him on his data and even suggested applying a filter in the settings when making the SL caps so the spectrographs could be viewed properly.

 

I called bullshit.

 

Hopefully we all can agree that backpedaling is not the right tactic at this point.

 

 

please stop telling people what they are agreeing with or not.  each person is capable of doing that for themselves.  :-)

 

i congratulated dd on bringing the topic to light, not for anything having to do with a pb13u subwoofer.

 

when folks measure the frequency response of their systems, a signal with log energy per frequency (equal energy per octave) is what is used.  even you do this because presenting a frequency response with equal energy per frequency would create a 3db per octave rise in the response and would have little interpretation. 

 

the point is simply that the same method should be used to present the content as was used to equalize the system.  otherwise, the lower frequencies will appear to rise at 3db/oct.

 

not much more to it than that.

[Electrodynamic]

please stop telling people what they are agreeing with or not.  each person is capable of doing that for themselves.  :-)

 

i congratulated dd on bringing the topic to light, not for anything having to do with a pb13u subwoofer.

 

when folks measure the frequency response of their systems, a signal with log energy per frequency (equal energy per octave) is what is used.  even you do this because presenting a frequency response with equal energy per frequency would create a 3db per octave rise in the response and would have little interpretation. 

 

the point is simply that the same method should be used to present the content as was used to equalize the system.  otherwise, the lower frequencies will appear to rise at 3db/oct.

 

not much more to it than that.

 

/end game. DIY measurements mean zero, but carry on. Good try Bosso. 

 

PS: The HST-11" DO woofer can not possibly crest 90 dB in-room regardless of power applied. Wait, what? Love you LTD 

 

...I remember when I was in my parent's basement posting theory instead of actual measured data. Although I do not rememer posting data based on computer modeling. 

[maxmercy]

This is why I don't go to AVS unless it is to see what PassingInterest is building, or Folgott is posting about.  We could use Folgott here, tremendous knowledge, implementation.  That riser he built is awesome.

 

JSS

[SME]

our hearing is log sensitive, so we use log signals (log sine sweeps, pink noise) to eq sound systems.  this way what appears on the screen and what we hear correlate.

More specifically, our perception of pitch varies logarithmically with frequency (to a very close approximation).  Hence, it is usually more useful to look at frequency on a log axis and to distribute measurements logarithmically across frequency.  That does not mean that our perception of loudness varies logarithmically with frequency, nor does it mean that systems are equalized to have any kind of 3 dB/octave response curve.

 

A properly equalized system will exhibit a flat response.  A flat response is one which doesn't substantially alter the signal, within its intended bandwidth.  A flat system fed white noise will output white noise.  A flat system fed pink noise will output pink noise.  A flat system fed an arbitrary signal, which could be represented as a superposition of sine waves, will output the same signal without altering the relative level of any of its constituent sine waves.

 

Furthermore, a system need not be equalized with pink noise.  One could also use white noise, or any other color of noise to equalize the system.  So long as the test signal covers the full bandwidth, and the system outputs the same signal that's input, the system is flat and is equalized properly.  Modern measurement systems use log sweeps because they have good signal-to-noise ratios and they make it very easy to separate THD from the linear part of the response.  However, provided that the test signal covers the full bandwidth, there's nothing stopping one from it for calibration.  Regardless of the method used, if the calibration is done correctly, then the resulting system will be flat.

 

log sine sweeps and pink noise have decreasing energy at 3db/oct.  as a result, spectrum lab, which presents data differently than the typical audio rta, presents such signals will have a 3db/oct downward slope.

 

as a result, in spectrum lab, when there is equal energy on the disc in the octave around 5hz, it will present 13db higher than the same amount of energy in the octave around 100hz (and another 10db higher than 1khz, and another 10db higher than 10khz, etc).

 

Both SpecLab and the RTA are functioning correctly here.  You are confused because they measure different things.  The RTA used for calibration actually measures power or energy per unit time within a set of fixed bins; whereas, Speclab measures power density or energy per unit time per unit frequency.  This difference is crucial because the power density at a particular frequency is independent of bin width; whereas the power measured in a particular bin on a calibration RTA depends on how wide the bin is.  In order to estimate the power density for frequencies within a bin with a given width and power, you must divide the power by the width.

 

A flat system will show approximately equal amounts of power from a pink noise source in the bands of an RTA if each band is half as wide as the band one octave higher.  As such, the RTA shows a biased response that cancels the bias in the pink noise test signal so that a flat system appears flat on the RTA with pink noise.

 

For assessing content, we want to know the level of signal at a particular frequency and at a particular time.  That's what the spectrogram is for.  The spectrogram yields energy density, which is numerically the same as the level if the dB scale is used.  This value does not depend on bin width as it does with the RTA.  As such, the spectrogram is a much more appropriate tool for assessing content than an RTA used for calibrations.

 

the tones that shredhead posted are showing equal energy per tone.  in spectrum lab, such content will appear to be at the same level.  we all agree on that, i hope.  of course the higher frequency tones will sound much louder, which is why we don't use this method to eq sound systems.  we all agree on that, i hope too.

 

But we do use this method to EQ systems.  A flat system will reproduce those sine waves at equal sound levels.  The fact that higher frequency tones sound louder than low frequency tones has no impact on how we equalize systems.  Differences in loudness perception are addressed by the mixer who will adjust the relative levels of frequencies in the mix until it sounds right.

 

the bottom line is that most folks who look at the spectrum lab captures are thinking that it is presenting data in the same way as the typical audio rta.  it is not.  the typical rta has the -3db/oct adjustment baked in, spectrum lab does not.  i'm not really sure that either could be called more or less "accurate", but one correlates with the way that we hear while the other does not.

 

I'm not sure what a typical RTA is, but I do know that RTAs designed for calibration are intentionally biased to read flat with a pink noise source.  This ensures that the end result will be flat when a pink noise source is used.  Such RTAs should never be used to assess content.  It's really that simple.  And no, the RTA does not correlate with what we hear at all.  Look at the Equal Loudness Contours and the Bark Scale Critical Bands and you will see absolutely no resemblance to the 3 dB/octave power spectrum of pink noise.

[LTD02]

thanks for taking the time to write your response sme.

 

my understanding is that pink noise is generally perceived as sounding 'flat' and that is why it is used and not some other signal for calibrating sound systems.  rta's calculate energy based on octaves (or portions thereof), which with pink noise, produce a visually flat response.  the end result being a pretty good correlation between what we see on the display and what we hear with our ears.  (i know its not perfect because of fletcher munson curves/equal loudness curves, etc., but close enough.)

 

now take that same pink noise content and move over to spectrum lab.  the pink noise still sounds flat, but the display shows a decline of 3db/oct.  across the audible bandwidth, this amounts to 30db or so and that doesn't correlate at all with what we hear.  so, to bring it back into alignment, a 3db/oct adjustment is necessary.  it is all simply an attempt to best correlate what is displayed visually with what we hear.  maybe there is a better way?