• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!

New set of graphs for all speakers reviews

pierre

Addicted to Fun and Learning
Forum Donor
Joined
Jul 1, 2017
Messages
962
Likes
3,048
Location
Switzerland
I had some time to kill in a plane and I played with all the data’s that Amir gives us.
you can see them on www.spinorama.org The code to generate them is On github Too.

just click on a graph to drill down, the last one is slightly interactive.
better contour plot and polar plot are under debugging. I will add a view to compare 2 speakers side by side.

if you like python notebook, you can easily play with the graphs yourself.

pull request welcome on github and please give me some feedback,
it is easy to improve.
Screenshot 2020-03-11 at 06.07.35.png
 
Last edited:

edechamps

Addicted to Fun and Learning
Forum Donor
Joined
Nov 21, 2018
Messages
910
Likes
3,620
Location
London, United Kingdom
Nice! I am also in the process of coding something quite similar but focusing on dissecting @MZKM's preference score calculations.

By the way, did you know you can use Colab for Jupyter notebooks? It makes it possible to use them completely in the browser without having to install anything. That's what I'm using right now for my little project. Here's a very basic example that prints the on-axis graph for the Micca RB42 after downloading the raw data from ASR on the fly.
 

sweetchaos

Major Contributor
The Curator
Joined
Nov 29, 2019
Messages
3,872
Likes
11,551
Location
BC, Canada
Hey @pierre
A couple of thoughts (in no particular order):
1. Set upper boundary to 20,000hz. No reason to go higher, as measurements only matter up to till 20,000hz.
2. You have 12 reviews, but there's a total of 15 so far.
3. Looking forward to the overlay ability, in order to compare 2 speakers at once.
4. Im looking at the "Gallery" section, and the graphs appear kind of small. Is there any way to increase the size to full screen, similar to @MZKM graphs?
5. While the list of speakers is small now, consider separating the active from passive ones into 2 folders on github, when the list gets bigger, in order to find models easier. @MZKM makes it clear which is active/passive.
6. While I was using the mouse to look at various measurements in dB, I couldn't see which exact Frequency (hz) that was for...I could only "eye-ball" it by looking at the bottom axis. Any way to include the Hz value on the mouse-over?

Great work so far!
 

hardisj

Major Contributor
Reviewer
Joined
Jul 18, 2019
Messages
2,907
Likes
13,908
Location
North Alabama
I really like what you've done here. I have been trying to find ways to have interactive charts for my site (link in sig) once I start testing again so this might help me with that. I use Matlab; no experience at all with Python. But maybe there's something here that I can use. So thanks again for sharing the code. :)

Using this set of data as my go-by, here are my comments.
https://pierreaubert.github.io/spinorama/Selah Audio RC3R.html

Spectrogram
  1. The spectrograms are very hard to read because they're pixelated. When I generate my spectrograms in Matlab they look like this by default. I just hide the edge lines (the white lines in your case) and also set the shading to interpolate. Nets me the pictures shown below. Maybe this link would be useful?
  2. I would also prefer the colors be what I've chosen below. Not because I said so... but because it more closely resembles what other programs use (VirtuixCAD, the Klippel plots, aRTA). That's why I chose the color scheme I did. Plus, the way the colors are now, it's honestly a bit confusing. For example, 0 dB and -9dB could both be gray/white. Which leads me to this...
  3. I would prefer to shorten the scaling from 0 to -60 to something like 0 to -30. Give finer gradients to highlight issues. I use maxSPL rounded to nearest +3 down to -30 and the number of color samples is the total bins divided by 3. That gets me color bins in 3dB increments.
  4. I noticed your highest value is 0. This implies the reference is whatever the max SPL of the spectro is on-axis. But there are spots off-axis where the SPL may be higher than the SPL at 0-degrees. Thanks to the color shading, it's hard to know for sure if that's how the plot is set up. I'm assuming each frequency is normalized against each frequency against the 0-degrees axis. But it's just not clear thanks to all the above.
  5. Set the x-axis to match the FR x-axis. Makes it easier to compare when jumping back and forth.
Please note: My spectro is not perfect so this is not me acting like I think it is. I've already updated it a couple times since I posted the one below. I'm just giving you some input based on some of my own trial and error and how it seems best to view this kind of data.


FR Plots:
  1. I second the suggestion above about having the frequency displayed. IMO, placing it at the bottom of the graph instead of at the FR trace would be less cluttered.
  2. Decimal places to 1/10 is adequate. No need for anything finer than that. The extra digits just clutter up the space.
  3. Can you add a zoom feature to the FR plots?
  4. Make the y-axis values of the FR plots consistent. I understand the CEA2034 being wider because you want to fit the DI curves in without overlapping the FR. But on the others, the y-axes between graphs are at different scales.
General feedback: I would personally prefer to see the graphics as a rectangle instead of a square. The longer x-axis is a bit easier to follow along and keeps frequency data from getting smushed. Just a personal choice.


I understand getting this stuff right isn't trivial. Like I said above, I'm not familiar with python code at all. Not sure how easy the suggestions are. So if it's just not feasible to do these things then that's fine.

Thanks for accepting feedback,
Erin


Here are the horizontal and vertical spectrograms from -90 to +90 degrees.

View attachment 49052

View attachment 49053
 
OP
P

pierre

Addicted to Fun and Learning
Forum Donor
Joined
Jul 1, 2017
Messages
962
Likes
3,048
Location
Switzerland
Hey @pierre
A couple of thoughts (in no particular order):
1. Set upper boundary to 20,000hz. No reason to go higher, as measurements only matter up to till 20,000hz.

The underlying library altair doesn't know how to do that. I will create a feature request to vega-light developers.

2. You have 12 reviews, but there's a total of 15 so far.
I will add the missing ones. I found a new one but didn't find the data for the first 3 reviews.

3. Looking forward to the overlay ability, in order to compare 2 speakers at once.
Code is done, i will integrated and push it tonight. Altair provides nice features for that.

4. Im looking at the "Gallery" section, and the graphs appear kind of small. Is there any way to increase the size to full screen, similar to @MZKM graphs?

Yes that's easy to do. I will had a javascript library to make it more responsive (to size of screen and kind of devices)

5. While the list of speakers is small now, consider separating the active from passive ones into 2 folders on github, when the list gets bigger, in order to find models easier. @MZKM makes it clear which is active/passive.
I created a feature request so i don't forget.

6. While I was using the mouse to look at various measurements in dB, I couldn't see which exact Frequency (hz) that was for...I could only "eye-ball" it by looking at the bottom axis. Any way to include the Hz value on the mouse-over?
I created another feature request so i don't forget, easy to do, will do it.

Great work so far!
Thanks for the feedback.
 
OP
P

pierre

Addicted to Fun and Learning
Forum Donor
Joined
Jul 1, 2017
Messages
962
Likes
3,048
Location
Switzerland
If you look at this graph [here], they are now interactive:

- you can zoom with the wheel, pan with the mouse.
- a tooltip popup when you move the mouse with better displayed informations.
- if you click on a legend (measurements) then 1 curve is highlighted, others are dimmed.
 
OP
P

pierre

Addicted to Fun and Learning
Forum Donor
Joined
Jul 1, 2017
Messages
962
Likes
3,048
Location
Switzerland
[QUOTE="hardisj]
FR Plots:
  1. I second the suggestion above about having the frequency displayed. IMO, placing it at the bottom of the graph instead of at the FR trace would be less cluttered.
  2. Decimal places to 1/10 is adequate. No need for anything finer than that. The extra digits just clutter up the space.
  3. Can you add a zoom feature to the FR plots?
  4. Make the y-axis values of the FR plots consistent. I understand the CEA2034 being wider because you want to fit the DI curves in without overlapping the FR. But on the others, the y-axes between graphs are at different scales.
[/QUOTE]

Hi Erin,

1. 2. and 3. are done.
4. is in the TODO list.

I will go after Spectogram soon.

Pierre
 
OP
P

pierre

Addicted to Fun and Learning
Forum Donor
Joined
Jul 1, 2017
Messages
962
Likes
3,048
Location
Switzerland
Hello,

work is crazy but i have done some improvements:

Data:
- i uploaded all data from ASR of course but also from Princeton/3d3a.
- if you have 10 minutes you can also use WebPlotDigitizer and scan a picture of a Spinorama

Website:
- website has a nice view with a search function, just type a brand name to select fewer speakers
- graphs are better looking, still some work to polish it.
- for each speaker: i have added the score from Olive paper. Some bugs are still there.

- I ranked all the speaker according to pref score (same as @MZKZ)
Screenshot 2020-03-11 at 05.52.15.png



- If you click on a speaker (in ASR or Princeton or Vendor), you will see a list of graphs. Click in a graph and you get an interactive view, with over etc...
- they are 2 other views I like: a vertical and a horizontal views with all graphs in 1. Here is an horizontal example. I want to generate a pdf from that but that's currently not done. I could generate a small book with all ASR measurements for example.

Screenshot 2020-03-11 at 05.57.37.png



I am playing with directivity metrics proposed by 3d3a, this one is freq v.s freq with each cell a normalize correlation between the horizontal and the vertical plots.

Screenshot 2020-03-11 at 06.00.28.png



I am learning, that's fun, that's why I do it. I put music on the genelec :) read some papers, do some experiment, nothing serious but
fun.

How you could help?
- digitilize some plots : I have experimented with a KEF reference 5 picture. Not great.
- help me figure out what's wrong with my PIR computation and friends (i do not have the same number as @MZKZ but I do not understand what's incorrect).
- various bugs: easy to track but i do not have the time right now. (For ex: some contour plot crash)
- propose features (no guarantee on delivery)

Pierre
 
OP
P

pierre

Addicted to Fun and Learning
Forum Donor
Joined
Jul 1, 2017
Messages
962
Likes
3,048
Location
Switzerland
Where is your computation?

I have found a few bugs. Scores now work as expected. Only remaining issues is on SP curve:
i still have an error increasing with frequency that i do not explain yet: that's for neumann speaker
control is what has been exported by the Klippel, computed are mine computation from
SPL H & V.

Code start here.

visualization (3).png
 

SDX-LV

Active Member
Joined
Jan 11, 2020
Messages
132
Likes
141
Location
Sweden
Hello,

work is crazy but i have done some improvements:

Data:
- i uploaded all data from ASR of course but also from Princeton/3d3a.
- if you have 10 minutes you can also use WebPlotDigitizer and scan a picture of a Spinorama

...

How you could help?
- digitilize some plots : I have experimented with a KEF reference 5 picture. Not great.
- help me figure out what's wrong with my PIR computation and friends (i do not have the same number as @MZKZ but I do not understand what's incorrect).
- various bugs: easy to track but i do not have the time right now. (For ex: some contour plot crash)
- propose features (no guarantee on delivery)

Pierre

Amazing work Pierre!

I am especially impressed about all the Princeton/3d3a data that you processed! - that data has been around for a while but you actually did the processing into Spinorama! THANK YOU!

Your GitHub page is extremely impressive. The WebPlotDigitizer "input" is an excellent addition - with it your database can absorb all Harman, Audioholics and a few other published Spinoramas into one database together with preference scores! You are very quickly making https://speakerdata2034.blogspot.com/2019/03/all-spinorama-data-index.html obsolete! Great work!

Some feedback at the moment:
  • Please force the X-axis to always go from 20Hz to 20kHz - right now some graphs are starting from 500Hz - messes up the comparison.
  • Right now the Vendor, digitized data pages do not work. So I did not check it.
  • Consider to add a comparison view - to see 2 Spinoramas overlayed over one another - this is the ultimate spin comparison approach and it seems to be within reach. :)
 
OP
P

pierre

Addicted to Fun and Learning
Forum Donor
Joined
Jul 1, 2017
Messages
962
Likes
3,048
Location
Switzerland
Amazing work Pierre!

I am especially impressed about all the Princeton/3d3a data that you processed! - that data has been around for a while but you actually did the processing into Spinorama! THANK YOU!

Your GitHub page is extremely impressive. The WebPlotDigitizer "input" is an excellent addition - with it your database can absorb all Harman, Audioholics and a few other published Spinoramas into one database together with preference scores! You are very quickly making https://speakerdata2034.blogspot.com/2019/03/all-spinorama-data-index.html obsolete! Great work!

Some feedback at the moment:
  • Please force the X-axis to always go from 20Hz to 20kHz - right now some graphs are starting from 500Hz - messes up the comparison.
  • Right now the Vendor, digitized data pages do not work. So I did not check it.
  • Consider to add a comparison view - to see 2 Spinoramas overlayed over one another - this is the ultimate spin comparison approach and it seems to be within reach. :)

Thanks.

- some graphs start at 500hz: Princeton 3d3a labs has a small anechoic room and data is not precise below 500hz which is why i removed them. 3d3a graphs also start at 500hz. they do have much better looking contour plot and i am learning how to generate nice contours: i discovered https://en.wikipedia.org/wiki/Marching_squares today and i will implement it shortly.
- comparison is coming too: it works in colab but i do not have enough space for free on github to host the generated files. One option is to host the file myself at home but my bandwidth is not great.
- vendors graphs are not working well, current interpolation is broken, import is painful. i am wondering if some machine learning could not give me much better results: i have a full set of graphs to learn on. I will experiment.
 
Last edited:
OP
P

pierre

Addicted to Fun and Learning
Forum Donor
Joined
Jul 1, 2017
Messages
962
Likes
3,048
Location
Switzerland
I scanned some spinorama for the big boys: not super precise but that gives an idea. Bass extension is better, so is the score.
This version is not yet available but coming real soon now :)

Screen Shot 2020-03-17 at 20.09.17.png
 

Haint

Senior Member
Joined
Jan 26, 2020
Messages
346
Likes
452
Hey Pierre, any tips on how to interpret the matrix graphs?
 
Last edited:
OP
P

pierre

Addicted to Fun and Learning
Forum Donor
Joined
Jul 1, 2017
Messages
962
Likes
3,048
Location
Switzerland
Hey Pierre, any tips on how to interpret the matrix graphs?

hello Haint,

i usually see where the crossover is easily for example. the other thing is that if color are very different between upper triangle and lower triangle, it means you have a lot of differences in directivity between horizontal and vertical planes.

That's more useful when they are normalised (as in a post above with 4 of them together). On the website, normalisation is the next feature.
 

stevenswall

Major Contributor
Forum Donor
Joined
Dec 10, 2019
Messages
1,366
Likes
1,075
Location
Orem, UT
I scanned some spinorama for the big boys: not super precise but that gives an idea. Bass extension is better, so is the score.

Does bass extension correlate with a higher score? If so, I really want to hear some larger speakers that go down to 20hz in room... Confirm my decisions and all that. ;)
 

StevenEleven

Addicted to Fun and Learning
Forum Donor
Joined
Dec 1, 2018
Messages
581
Likes
1,188
This is truly wonderful work.

My only concern is, based on what I have seen here at ASR, I find the manufacturer data from Harmon and KEF to be not credible at all. I’d ditch those until we get some reliable independent data from @amirm or someone else. The KEF and Harman data in my view is simply not to be taken at face value. So I’d say either leave those out or put an asterisk next to the calculations based on KEF or JBL/Harman published data and reference it as being based on vendor data that is not independently verified or independently corroborated. :)
 
Last edited:
Top Bottom