spinorama.org

[richard12511]

I don’t have much doubt. Though the Stereophile measurements show a bump around 1800Hz:https://www.stereophile.com/content/kef-reference-5-loudspeaker-measurements


The issue is the formula liking narrow directivity, with the Reference 5 is, so with real human trials it likely wouldn’t score as high as the formula predicts.

It seems odd that the formula favors narrow directivity. The listening tests heavily favored wide directivity, so much so that Toole lists wide dispersion as the second most important metric after flat LW.

 

[MZKM]

It seems odd that the formula favors narrow directivity. The listening tests heavily favored wide directivity, so much so that Toole lists wide dispersion as the second most important metric after flat LW.

It’s not meant to. Olive wanted something to analyze the PIR curve, but it’s sloping down, so applying NBD (deviation per band) wouldn’t be as good as of course there would be deviation as it’s not flat. Smoothness (correlation in stats) looks for a linear relationship, and any deviation from a line (regardless of slope) gets dinged, so it seems like a good fit, plus it doesn’t ding deviations in treble as much as in midrange, so another plus. The issue is, the more slopped a line is, the stronger the relationship between frequency and dB (the more frequency increases the lower the dB), thus 2 speakers could have the same amount of deviation at the same frequencies, but one that is narrower dispersion will get a higher score.

 

[jonfitch]

It seems odd that the formula favors narrow directivity. The listening tests heavily favored wide directivity, so much so that Toole lists wide dispersion as the second most important metric after flat LW.

I wonder how room size effects this. I recall photos in the marketing literature on the listening tests that looked like mono speaker testing in theater-sized rooms.

Question is also how much would you still care about wide directivity once you have 2 speakers. Because a lot of listeners are preconditioned to hear sound from left and right sources. When you listen to a single speaker you may want something more analogous to that and thus prefer wider directivity.

 

[richard12511]

I wonder how room size effects this. I recall photos in the marketing literature on the listening tests that looked like mono speaker testing in theater-sized rooms.

Question is also how much would you still care about wide directivity once you have 2 speakers. Because a lot of listeners are preconditioned to hear sound from left and right sources. When you listen to a single speaker you may want something more analogous to that and thus prefer wider directivity.

I'm conflicted on this. Traditional Toole science says that the same speakers are preferred in all rooms. This has not been my experience, though.

IME, wide dispersion speakers (like Revel) tend to sound better when the distance to the first side wall reflection is larger. Narrower dispersion speakers(like KEF) tend to sound better when the distance to the first side wall reflection is close. KEF seems to think this too, as I've read the reason they go for more narrow dispersion is that they sell more to the European market, which tends to have smaller rooms.

 

[pierre]

That Kef Reference 5 sits uncomfortably on top of the rankings lol. I wonder when we will get a real verified 3rd party spin vs just Kef's published charts.

Hello,

I wanted to test for a long time how the score would change when data are noisier or smoother. I made a quick experiment:

def noise(df, amp):
    sdf = df.copy()
    l = len(sdf.Freq.values)
    for c in sdf.columns:
        if c != 'Freq':
            sdf[c] += amp*(np.random.rand(l)-0.5)
    return sdf

def smooth(df, amp):
    sdf = df.copy()
    l = len(sdf.Freq.values)
    for c in sdf.columns:
        if c != 'Freq':
            sdf[c] = sdf[c].rolling(amp).mean()
    sdf.fillna(df)
    return sdf

and to compute the modified scores:

def eval_score_noise(df_all, speaker, amplitude):
    df_klippel = df_all[speaker]['ASR']['asr']
    spl_H = noise(df_klippel['SPL Horizontal_unmelted'], amplitude)
    spl_V = noise(df_klippel['SPL Vertical_unmelted'], amplitude)
    df_smoothed = load_normalize(filter_graphs('Adam S2V', spl_H, spl_V))
    return scores(df_smoothed)['pref_score']

def eval_score_smooth(df_all, speaker, amplitude):
    df_klippel = df_all[speaker]['ASR']['asr']
    spl_H = smooth(df_klippel['SPL Horizontal_unmelted'], amplitude)
    spl_V = smooth(df_klippel['SPL Vertical_unmelted'], amplitude)
    df_smoothed = load_normalize(filter_graphs('Adam S2V', spl_H, spl_V))
    return scores(df_smoothed)['pref_score']

which yields the following results for the first tested speaker Adam S2V:

Noise +/-0.0dB score 5.5
Noise +/-0.3dB score 5.5
Noise +/-0.7dB score 5.5
Noise +/-1.0dB score 5.4
Noise +/-1.3dB score 5.3
Noise +/-1.7dB score 5.3
Noise +/-2.0dB score 5.2
Noise +/-2.3dB score 4.9
Noise +/-2.7dB score 4.7
Noise +/-3.0dB score 4.8

as expected the score decreases with the noise.

Size of rolling window # 2 score 5.8
Size of rolling window # 3 score 6.0
Size of rolling window # 4 score 6.2
Size of rolling window # 5 score 6.2
Size of rolling window # 6 score 6.3
Size of rolling window # 7 score 6.3
Size of rolling window # 8 score 6.3
Size of rolling window # 9 score 6.4
Size of rolling window #10 score 6.4
Size of rolling window #20 score 6.4

and also expected, the score increases with the smoothing.

A bit more complicated is how to estimate the amount of smoothing of graphs which have been digitalized.

Full code https://github.com/pierreaubert/spinorama/blob/develop/experimental/play_smoothing.ipynb

 

[jonfitch]

Interesting, so if I'm reading right probability is it will still be in the mid to high 6 range it seems, probably still higher than Salon 2.

 

[sweetchaos]

Fantastic work on keeping this project up-to-date, @pierre !
I'm using it more and more every day!

I noticed a few discrepancies on your github page (link), so I thought I should mention them:
1. When I selected "ErinsAudioCorner" from the dropdown, I don't see Erin's "Kef R3" review.
I do see it when all the options are at default, though.
2. Missing Erin's "Klipsch Heresy IV" review
3. Missing Erin's "Micca MB42X Mark III" review
4. Missing Erin's "Neumi BS5" review
5. Missing Erin's "Jamo S807" review.
6. Missing Erin's "Buchardt Audio S400" review
7. I didn't see any spins from napilopez? He's made about 15 or so.
just noticed something else: 8. the picture for "IK Multimedia iLoud MTM" is currently wrong, as it's showing "IK Multimedia iLoud Micro Monitor"

 

[pierre]

Fantastic work on keeping this project up-to-date, @pierre !
I'm using it more and more every day!

I noticed a few discrepancies on your github page (link), so I thought I should mention them:
1. When I selected "ErinsAudioCorner" from the dropdown, I don't see Erin's "Kef R3" review.
I do see it when all the options are at default, though.
2. Missing Erin's "Klipsch Heresy IV" review
3. Missing Erin's "Micca MB42X Mark III" review
4. Missing Erin's "Neumi BS5" review
5. Missing Erin's "Jamo S807" review.
6. Missing Erin's "Buchardt Audio S400" review
7. I didn't see any spins from napilopez? He's made about 15 or so.
just noticed something else: 8. the picture for "IK Multimedia iLoud MTM" is currently wrong, as it's showing "IK Multimedia iLoud Micro Monitor"

1. looks like a bug, I will look at it. The javascript part of the project has been hacked quickly and doesn't have the quality of the python part yet.

For 2 to 7, I think the data has not been released (only the pictures).
Here is how you can help:
- petition @hardisj and @napilopez to release the data
- failing that, take the hard boring path and digitalise the picture with webplotigitizer.

I am spending my little available free time to build a model for loudspeakers:
- currently I have a loudspeaker modelled a set of drivers+crossovers.
- i want to use it like this:
- model
- add noise or room impact on the model
- apply room correction
- check that you are close to the model

Too many projects, so little time.

 

[sweetchaos]

@hardisj and @napilopez
We need your help to release the raw data for your spinorama graphs, such that @pierre and @MZKM and @Maiky76 can work their magic, and create wonderful graphs, and EQ profiles for everyone to enjoy!

 

[napilopez]

1. looks like a bug, I will look at it. The javascript part of the project has been hacked quickly and doesn't have the quality of the python part yet.

For 2 to 7, I think the data has not been released (only the pictures).
Here is how you can help:
- petition @hardisj and @napilopez to release the data
- failing that, take the hard boring path and digitalise the picture with webplotigitizer.

I am spending my little available free time to build a model for loudspeakers:
- currently I have a loudspeaker modelled a set of drivers+crossovers.
- i want to use it like this:
- model
- add noise or room impact on the model
- apply room correction
- check that you are close to the model

Too many projects, so little time.

@hardisj and @napilopez
We need your help to release the raw data for your spinorama graphs, such that @pierre and @MZKM and @Maiky76 can work their magic, and create wonderful graphs, and EQ profiles for everyone to enjoy!

Ive been reluctant to release my data because there's a fair bit more jerry-rigging going on with my measurements, I'm constantly refining my setup, and I didn't want people to be calculating preference scores to compare with results for the NFS.

For example, sometimes there are small dips on the data that are from my measurement setup rather than the speaker itself, and I usually point these out. Likewise while I merge nearfield and garfield data to create spins, I tend to prefer sharing the off-axis data in it's pure, unspliced form.

But I suppose the data is out there anyway in photo form, so I'm happy to at least share the completed spin files.

 

[hardisj]

Here is how you can help:
- petition @hardisj and @napilopez to release the data

All my NFS data has been released. The Kef R3 data was posted in that thread, but buried somewhere in there.... though, I thought I had given you the link already. At any rate, here it is again:
https://www.dropbox.com/s/npddybx4v06ytps/Kef R3.7z?dl=0


The other data from the other tests were conducted before I had the NFS and aren't in a format suitable for your tool and aren't worth my time to put in to a proper format. That said, the Klipsch Heresy IV was re-measured with the NFS and I will be sharing that data once I create the review.

 

[pierre]

All my NFS data has been released. The Kef R3 data was posted in that thread, but buried somewhere in there.... though, I thought I had given you the link already. At any rate, here it is again:
https://www.dropbox.com/s/npddybx4v06ytps/Kef R3.7z?dl=0


The other data from the other tests were conducted before I had the NFS and aren't in a format suitable for your tool and aren't worth my time to put in to a proper format. That said, the Klipsch Heresy IV was re-measured with the NFS and I will be sharing that data once I create the review.

if you have spl horizontal and vertical (like the Bose) that works too. I very much understand if that’s too much work to do.
Dont’t worry. Keep your time for running the NFS on new speakers.

 

[sweetchaos]

@napilopez
That's because you're a perfectionist. I see this differently...I take the view of having something is better than nothing. "Done is better than perfect". You've given us invaluable spinorama data. Can we expect it to be the same quality of NFS or anechoic data? Hell no. Who would be thinking that?! Haha

In my "subwoofer comparison" spreadsheet, I'm including several "other" sources of CEA-2010-A in my, other than Erin/Data-bass/Audioholics/BrentB. Do I know how well these "other" sources are preforming the measurements to the same degree of professional care as Erin/Data-bass/Audioholics/BrentB? No. But I don't mind. I'm simply grateful that they shared it. I do my part and always link back to their article.

Same for you. As long as @pierre puts a link back to your article (where you've carefully explained your data ), you're set. Anyone who expects to compare your measurements to NFS/anechoic data, and expect 1:1 results is fooling themselves, so you have nothing to worry about.

 

[sweetchaos]

Which brings me to an area of improvement for @pierre
I think you should add the following to your project:
1. The link to original review = which you've already done
2. Type of measurements = NFS, anechoic, semi-anechoic, quasi-anechoic, etc.
3. Source of data. Raw (provided by source) or Digitized (example, using WebPlogDigitizer to extract data from posted graph)

I can see that 2 reviews by TimVG were included. Both were done using quasi-anechoic measurements and their raw data was released to make the spinorama graphs.
I can also see that several spinorama's were included from various vendors, and I believe some of these were done using digitizing of the original graph.
I'm saying it would be nice to understand that difference when looking at plots on your project.

Also, I have no issues with digitizing plots for Erin and a few others that you've missed. Meaning, I'll help you out in that regard, seeing how your time is also limited right now.

 

[sweetchaos]

@hardisj thanks for your feedback.

In that case, I'll digitize these 5 (non-NFS) graphs in the meanwhile, and so pierre can add these speakers to his project.

Great news with re-measuring the Klipsch Heresy IV on the NFS! That gives us more accurate data, I love it.

 

[Beershaun]

hello,

you have 2 options: first is to take the filters in order (the first 10). second option is to use the software and generates the EQ you want:

./generate_peqs.py --help

will give you the list of options. Here is an example with the JBL HDI-4500

./generate_peqs.py --speaker='JBL HDI-4500' --slope-listening-window=-0.5 --max-Q=8 --max-peq=10 --force

which gives

EQ for JBL HDI-4500 computed from ASR data
Preference Score 3.6 with EQ 5.3
Generated from http://github.com/pierreaubert/spinorama/generate_peqs.py v0.7
Dated: 2021-04-05-10:03:43
Preamp: -2.7 dB
Filter  1: ON PK Fc   546 Hz Gain +3.13 dB Q 3.41
Filter  2: ON PK Fc 14427 Hz Gain -4.34 dB Q 2.02
Filter  3: ON PK Fc  1914 Hz Gain +2.33 dB Q 5.49
Filter  4: ON PK Fc 10393 Hz Gain +2.66 dB Q 8.00
Filter  5: ON PK Fc  6120 Hz Gain -1.11 dB Q 1.00
Filter  6: ON PK Fc  1563 Hz Gain +1.17 dB Q 8.00
Filter  7: ON PK Fc   604 Hz Gain -1.19 dB Q 8.00
Filter  8: ON PK Fc   456 Hz Gain -0.78 dB Q 8.00
Filter  9: ON PK Fc   693 Hz Gain +1.04 dB Q 2.12
Filter 10: ON PK Fc  3002 Hz Gain +0.92 dB Q 8.00

Hi @pierre
I'm having some difficulty setting up the code to run on my local Mac. Where would you like me to post error logs and questions?

 

[pierre]

Hi @pierre
I'm having some difficulty setting up the code to run on my local Mac. Where would you like me to post error logs and questions?

Maybe easier to follow on GitHub. Open a support request.

 

[sweetchaos]

Well, I did what I said I would...

I've created a dropbox, so I can drop all the WebPlotDigitizer files inside, and not flood the forum with attachments...
https://www.dropbox.com/sh/3wi5ccj89b8v9iw/AADuprmL5-9esq0CJ8mEq5y2a?dl=0

DONE:
1. Neumann KH80 - from Neumann, posted here on ASR
2. Neumann KH120 - from Neumann, posted here on ASR
3. Klipsch Heresy IV - from Erin
4. Micca MB42X Mark III - from Erin
5. Neumi BS5 - from Erin
6. Jamo S807 - from Erin
7. Buchardt Audio S400 - from Erin
8. JBL L82 Classic - from vendor
9. JBL 4312SE - from vendor
10. Kef Reference 5 prototype 2014 - from vendor
11. JBL LSR6332 - from vendor
12. Bowers & Wilkins 803 D3 - from vendor
13. Infinity R162 - from vendor
14. Infinity R263 - from vendor

TO-DO:
Archimago's Musings
- Emotiva Airmotiv B1+
- Spendor SA1 (1976)
- Radio Shack / Realistic / Tandy Minimus 7 speakers (early 1980s, Cat. No. 40-2030A)
- ELAC Debut 2.0 A4.2 Atmos Speakers
- Tannoy REVEAL 501a
- Fluance Reference XL8S
- AudioEngine A2
- Edifier S2000 Mk III
- KEF LS50
Audioholics maybe?
-There's over 20+ I think.

 

[pierre]

Well, I did what I said I would...

I've created a dropbox, so I can drop all the WebPlotDigitizer files inside, and not flood the forum with attachments...
https://www.dropbox.com/sh/3wi5ccj89b8v9iw/AADuprmL5-9esq0CJ8mEq5y2a?dl=0

DONE:
1. Neumann KH80 - from Neumann, posted here on ASR
2. Neumann KH120 - from Neumann, posted here on ASR
3. Klipsch Heresy IV - from Erin
4. Micca MB42X Mark III - from Erin
5. Neumi BS5 - from Erin
6. Jamo S807 - from Erin
7. Buchardt Audio S400 - from Erin
8. JBL L82 Classic - from vendor
9. JBL 4312SE - from vendor
10. Kef Reference 5 prototype 2014 - from vendor
11. JBL LSR6332 - from vendor
12. Bowers & Wilkins 803 D3 - from vendor
13. Infinity R162 - from vendor
14. Infinity R263 - from vendor

TO-DO:
Archimago's Musings
- Emotiva Airmotiv B1+
- Spendor SA1 (1976)
- Radio Shack / Realistic / Tandy Minimus 7 speakers (early 1980s, Cat. No. 40-2030A)
- ELAC Debut 2.0 A4.2 Atmos Speakers
- Tannoy REVEAL 501a
- Fluance Reference XL8S
- AudioEngine A2
- Edifier S2000 Mk III
- KEF LS50
Audioholics maybe?
-There's over 20+ I think.

Great work. The parser did find a few mistakes, I have modified the code to handle them since they are very common and I have tripped myself in the same way in the past.

The KEF Ref 5 I already have (only known measurement I think).

Website is updating...

 

[napilopez]

@napilopez
That's because you're a perfectionist. I see this differently...I take the view of having something is better than nothing. "Done is better than perfect". You've given us invaluable spinorama data. Can we expect it to be the same quality of NFS or anechoic data? Hell no. Who would be thinking that?! Haha

In my "subwoofer comparison" spreadsheet, I'm including several "other" sources of CEA-2010-A in my, other than Erin/Data-bass/Audioholics/BrentB. Do I know how well these "other" sources are preforming the measurements to the same degree of professional care as Erin/Data-bass/Audioholics/BrentB? No. But I don't mind. I'm simply grateful that they shared it. I do my part and always link back to their article.

Same for you. As long as @pierre puts a link back to your article (where you've carefully explained your data ), you're set. Anyone who expects to compare your measurements to NFS/anechoic data, and expect 1:1 results is fooling themselves, so you have nothing to worry about.

I would definitely feel a lot more comfortable with it if pierre took your suggestion of differentiating the measurement type, e.g. anechoic, NFS, quasi-anechoic =]

People down the line won't know the full history of the measurements or necessarily be following closely the events of this forum to know how the measurements were taken and potential limitations.

In any case, I'll be including the spin curves from here on out. I'll consider doing individual angles too.