• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required as is 20 years of participation in forums (not all true). Come here to have fun, be ready to be teased and not take online life too seriously. We now measure and review equipment for free! Click here for details.

Dali Spektor 1 Review (Bookshelf Speaker)

jeando

New Member
Joined
Jul 19, 2021
Messages
3
Likes
0
With PW link correction near wall on desktop
1626712602527.png
 

cbracer

Member
Forum Donor
Joined
Jun 13, 2021
Messages
26
Likes
11
Location
California
I've been reading this forum for a while now, and not only learned a lot but excited when it reinforces some of my own findings. It appears if Amir doesn't put a speaker on his "yes" recommendation then many forum readers think the speaker is crap and use that as their justification. But with some speakers, like this one, Amir is more neutral saying he doesn't love it, it has some issues, but it isn't horrible. So I take that as a neutral rating. And for good reason this speaker's data results are actually quite good in some ways. Seems like people want to point to the negatives without discussing the positives. The directivity is a solid 60 degrees, and that goes further up the frequency range than most speakers. Plus the vertical directivity above the tweeter is really good compared to other speakers. Also, trying to measure noise at 96db for every speaker may not fair because some speakers (small) aren't designed for that while others are. What would be more useful is data at each +1 or +2 db range up to find out when the noise floor really does begin to move up. A speaker that gets distorted at 89db would have the same rating as one that gets distorted at 95db because we only have data at 86 and 96. If we could say around XX value is where the noise starts to get out of control, much like we see with testing amplifiers, that would be useful info in reviews.
 

modes

Member
Joined
Apr 6, 2019
Messages
24
Likes
54
They were on sale for 140€ a few years back so I decided to hoard a pair. Subjectively I'd say I enjoy them more than some of my far more expensive, larger bookshelves from certain British and French mf's.

Fun fact: they really skimped on the crossover. Partly 1st order (and the tiniest, nastiest bipolars in the business) Now that I think of it, the resulting frequency/amplitude graph is killin' it, considering.
1626718596142.jpg
 
Last edited:

uwotm8

Member
Joined
Jul 14, 2020
Messages
97
Likes
73
In Soviet Russia price for pair pair of these is roughly $180 same as Wharfedale D310.
Both are the cheapest hi-fi speakers if you don't mind some noname garbage.
I'd call Spektor performance not great, not terrible:p
 

uwotm8

Member
Joined
Jul 14, 2020
Messages
97
Likes
73
They were on sale for 140€ a few years back so I decided to hoard a pair. Subjectively I'd say I enjoy them more than some of my far more expensive, larger bookshelves from certain British and French mf's.

Fun fact: they really skimped on the crossover. Partly 1st order (and the tiniest, nastiest bipolars in the business) Now that I think of it, the resulting frequency/amplitude graph is killin' it, considering.
View attachment 142248
I'd definitely change HF cap to chinese red MKP
 

kimmosto

Member
Joined
Sep 12, 2019
Messages
65
Likes
101
Preference Rating
SCORE: 3.1
SCORE w/ sub: 5.7

I'm testing possibilities to add calculation of preference rating to VituixCAD, but everything is not clear for me. So few questions:

1) How do you calculate score with (ideal) sub? Do you set LFX=0 (and LFQ=0 with equation 10) or some other small constant?

2) How do you calculate SM? Patent application says that "SM is the Pearson correlation coefficient of determination (r^2)..." (such as RSQ function in Excel). Problem is that r^2 is closing to 0 if slope of Y is zero so for example SM_ON is very small (=worst possible) with flat horizontal response. That does not make sense because flat is typical target for on-axis. This causes problems with both Equation 9 and 10 including at least one SM.
Using regression line as X in SM calculation does not change anything because slope of Y is still close to zero with flat responses.
Mean squared error to linear regression works much better, but result needs some scaling e.g. 1/(1+err/2.5) to be more compatible with r^2 for tilting curves such as PIR and SP. Another cheap trick would be tilting SPL responses down at HF to get higher correlation also for on-axis.

In addition, flat on-axis target is missing in Equation 9 so optimizing crossover with preference rating could create total disaster - for example by attenuating LW at HF to get LFX down -> Y1 up.

One more comment about LFX. Patent application suggests SP for LFX and LFQ calculation which looks assumption that bass radiators are omni-directional. Not for example dipole or cardioid with DI~5 dB. Both LFX_SP and LFX_LW (and LFQ_SP and LFQ_LW) should be calculated, and smaller one selected for Y1 calculation in order to support DI >> 0 below 300 Hz.
 
Last edited:

kimmosto

Member
Joined
Sep 12, 2019
Messages
65
Likes
101
Sean Olive, would you like to comment/answer questions about SM calculation, on-axis tilt with Equation 9 and LFX/LFQ calculation for directive LF radiators?

Any plans to upgrade model for predicted preference rating to improve correlation and support for different speakers concepts such as directive LF and LF tilt? How about adding variables (other than amplitude responses) such as total directivity, compression spectrum and timing?
 
Last edited:

modes

Member
Joined
Apr 6, 2019
Messages
24
Likes
54
Spektor 1 replaced the Zensor 1, but they look cheaper, and it looks like they perform worse as well (Zensor is quite flat off axis, and like all Dali speakers, they are not designed for on axis listening).
Not sure if you can call it a direct replacement. Here the spektor 1 costs ~38% lower and has a 4" woofer instead of the Zensor 1's 5".
 

Colonel7

Senior Member
Joined
Feb 22, 2020
Messages
412
Likes
518
Location
Maryland, USA
I'm testing possibilities to add calculation of preference rating to VituixCAD, but everything is not clear for me. So few questions:

1) How do you calculate score with (ideal) sub? Do you set LFX=0 (and LFQ=0 with equation 10) or some other small constant?

2) How do you calculate SM? Patent application says that "SM is the Pearson correlation coefficient of determination (r^2)..." (such as RSQ function in Excel). Problem is that r^2 is closing to 0 if slope of Y is zero so for example SM_ON is very small (=worst possible) with flat horizontal response. That does not make sense because flat is typical target for on-axis. This causes problems with both Equation 9 and 10 including at least one SM.
Using regression line as X in SM calculation does not change anything because slope of Y is still close to zero with flat responses.
Mean squared error to linear regression works much better, but result needs some scaling e.g. 1/(1+err/2.5) to be more compatible with r^2 for tilting curves such as PIR and SP. Another cheap trick would be tilting SPL responses down at HF to get higher correlation also for on-axis.

In addition, flat on-axis target is missing in Equation 9 so optimizing crossover with preference rating could create total disaster - for example by attenuating LW at HF to get LFX down -> Y1 up.

One more comment about LFX. Patent application suggests SP for LFX and LFQ calculation which looks assumption that bass radiators are omni-directional. Not for example dipole or cardioid with DI~5 dB. Both LFX_SP and LFX_LW (and LFQ_SP and LFQ_LW) should be calculated, and smaller one selected for Y1 calculation in order to support DI >> 0 below 300 Hz.
Calling @MZKM
 

MZKM

Major Contributor
Forum Donor
Joined
Dec 1, 2018
Messages
3,866
Likes
9,643
Location
Land O’ Lakes, FL
I'm testing possibilities to add calculation of preference rating to VituixCAD, but everything is not clear for me. So few questions:

1) How do you calculate score with (ideal) sub? Do you set LFX=0 (and LFQ=0 with equation 10) or some other small constant?

2) How do you calculate SM? Patent application says that "SM is the Pearson correlation coefficient of determination (r^2)..." (such as RSQ function in Excel). Problem is that r^2 is closing to 0 if slope of Y is zero so for example SM_ON is very small (=worst possible) with flat horizontal response. That does not make sense because flat is typical target for on-axis. This causes problems with both Equation 9 and 10 including at least one SM.
Using regression line as X in SM calculation does not change anything because slope of Y is still close to zero with flat responses.
Mean squared error to linear regression works much better, but result needs some scaling e.g. 1/(1+err/2.5) to be more compatible with r^2 for tilting curves such as PIR and SP. Another cheap trick would be tilting SPL responses down at HF to get higher correlation also for on-axis.

In addition, flat on-axis target is missing in Equation 9 so optimizing crossover with preference rating could create total disaster - for example by attenuating LW at HF to get LFX down -> Y1 up.

One more comment about LFX. Patent application suggests SP for LFX and LFQ calculation which looks assumption that bass radiators are omni-directional. Not for example dipole or cardioid with DI~5 dB. Both LFX_SP and LFX_LW (and LFQ_SP and LFQ_LW) should be calculated, and smaller one selected for Y1 calculation in order to support DI >> 0 below 300 Hz.
1) For score with sub, I overwrite the actual LFX extension and give it a fixed extension which is I think between 14 & 15 Hz; I chose this as it's what I computed to give a score of 10 if all other parameters where at max (even though theoretically impossible). Some other people calculating a similar score simply use 20Hz, some have argued my method inflates the score too much; but it's never been tested (and we still don't have anything >9) so I don't really care.

2) It is simply r^2 of the trend line when log-scaled. You don't use SM for on-axis for the formula (you can see in the patent that Olive did use it as well as others to see which aspects were most important).


The largest issue is with the calculation of NBD, as it says to make 1/2 octave bands from100Hz-12kHz, but that doesn't work out evenly if those are the start/end frequencies, so I simply worked with it until it became the included frequencies, I use 14 1/2 octave bands starting at 96Hz and end at 12270Hz. I have asked Sean Olive about this but he was never able to get back to me.
 
Last edited:

kimmosto

Member
Joined
Sep 12, 2019
Messages
65
Likes
101
^Thanks MZKM! This is off-topic in Dali Spektor 1 thread, but short comments:
I already added calculation and maximizing (with Optimizer) of predicted preference rating to VituixCAD. Differences to patent application by Olive are listed in user manual. Both equations (9 and 10) visible in the patent are available for users.
* Equation 9 (which ASR uses) does not include requirement for on-axis tilt. Tilt can be almost anything so bad sound balance can produce high score if NBDs, LFX and SM_PIR are good. Negative tilt (of LW) produces higher LFX which could increase the problem with automatic iteration. Equation 9 is available for comparison to e.g. ASR, but cannot be recommended for design purposes.
* r^2 is sensitive to absolute value of slope. Certain (usually higher) slopes give higher correlation than slopes close to zero or infinite. r^2 as SM is replaced with scaled 'mean squared error to linear regression' in VituixCAD because equation 10 includes also SM_ON (and SM_SP). Equation 10 includes also AAD_ON so requirement for on-axis tilt is secured, but I prefer logical and stable variables anyway.
* NBD is calculated with "moving" smoothing of 1/2 oct with 6th order Butterworth BP slope because it's available and continuously symmetrical i.e. does not create any steps or variable weights within each band. I have not investigated how much difference this causes compared to calculation with fixed 1/2 oct bands.
* LFX and LFQ 'with sub' are set equal to But18 HP at 17.5 Hz limiting score to 10.0 with equation 9 (max with equation 10 is higher).
* Smoothing for NBD (and AAD) is 1/33 oct because VituixCAD already has internal 1/48 oct. scale. Measurements at home with time window are usually smoothed at mid-range so preference ratings are probably higher while XO simulation than for final speaker measured in large anechoic or with NFS.
* LFX calculation is able to select LW instead of SP for directive LF radiators such as dipoles and cardioids. SP alone is not universal enough for detecting -6 dB point relative to LW at mid...high.

Main/official VituixCAD thread is on HTguide where we can continue if necessary.
 
Last edited:
OP
amirm

amirm

Founder/Admin
Staff Member
CFO (Chief Fun Officer)
Joined
Feb 13, 2016
Messages
35,320
Likes
134,794
Location
Seattle Area

kimmosto

Member
Joined
Sep 12, 2019
Messages
65
Likes
101
^Table at Speaker Review and Measurement Index is able to sort records with Score and Score /w sub so ASR uses results calculated with equation 9. You better go to sleep and stop nitpicking if nothing smarter to tell us.
 
OP
amirm

amirm

Founder/Admin
Staff Member
CFO (Chief Fun Officer)
Joined
Feb 13, 2016
Messages
35,320
Likes
134,794
Location
Seattle Area
^Table at Speaker Review and Measurement Index is able to sort records with Score and Score /w sub so ASR uses results calculated with equation 9. You better go to sleep and stop nitpicking if nothing smarter to tell us.
The index is maintained by members, and has even reviews performed by others. It is not an ASR thing. None of my reviews have any preference scores.

One more obnoxious and rude remark from you and you will be banned.
 

MZKM

Major Contributor
Forum Donor
Joined
Dec 1, 2018
Messages
3,866
Likes
9,643
Location
Land O’ Lakes, FL
^Thanks MZKM! This is off-topic in Dali Spektor 1 thread, but short comments:
I already added calculation and maximizing (with Optimizer) of predicted preference rating to VituixCAD. Differences to patent application by Olive are listed in user manual. Both equations (9 and 10) visible in the patent are available for users.
* Equation 9 (which ASR uses) does not include requirement for on-axis tilt. Tilt can be almost anything so bad sound balance can produce high score if NBDs, LFX and SM_PIR are good. Negative tilt (of LW) produces higher LFX which could increase the problem with automatic iteration. Equation 9 is available for comparison to e.g. ASR, but cannot be recommended for design purposes.
* r^2 is sensitive to absolute value of slope. Certain (usually higher) slopes give higher correlation than slopes close to zero or infinite. r^2 as SM is replaced with scaled 'mean squared error to linear regression' in VituixCAD because equation 10 includes also SM_ON (and SM_SP). Equation 10 includes also AAD_ON so requirement for on-axis tilt is secured, but I prefer logical and stable variables anyway.
* NBD is calculated with "moving" smoothing of 1/2 oct with 6th order Butterworth BP slope because it's available and continuously symmetrical i.e. does not create any steps or variable weights within each band. I have not investigated how much difference this causes compared to calculation with fixed 1/2 oct bands.
* LFX and LFQ 'with sub' are set equal to But18 HP at 17.5 Hz limiting score to 10.0 with equation 9 (max with equation 10 is higher).
* Smoothing for NBD (and AAD) is 1/33 oct because VituixCAD already has internal 1/48 oct. scale. Measurements at home with time window are usually smoothed at mid-range so preference ratings are probably higher while XO simulation than for final speaker measured in large anechoic or with NFS.
* LFX calculation is able to select LW instead of SP for directive LF radiators such as dipoles and cardioids. SP alone is not universal enough for detecting -6 dB point relative to LW at mid...high.

Main/official VituixCAD thread is on HTguide where we can continue if necessary.
Equation 10 isn't used by any of us (as Olive found it didn't work will with the larger set), so any items that could be improved haven't been looked at too much.

As for use RMSE instead of r^2, that will have to be evaluated to see if it is more accurate. In my testing of a new formula I simply normalized the curve to it's own slope and calculated NDB & AAD on it.
To my understanding, r^2 is more lenient on treble deviation, is the same true for RMSE?

As for use LW instead of SP for LFX of unique speakers; that seems good but will also have to be investigated with real in-room measurements. Take the Magnepan LRS for instance:
index.php


LW for sure has more bass than SP & ER & PIR, but which one is more accurate for showing bass at the listening position? Since bass is not omni, that means less energy in the room (but the Spinorama does not take into account boundary gain, which the LRS will have a lot of as the driver goes down to the floor).

Any further talk probably should be in my thread and not in this speaker review.
https://www.audiosciencereview.com/...erence-ratings-for-loudspeakers.11091/page-28
 

restorer-john

Master Contributor
Joined
Mar 1, 2018
Messages
7,818
Likes
20,841
Location
Gold Coast, Queensland, Australia
I've asked removing my username already twice so just do it and don't write any threats.

Can't we all just play nice? We aren't saving the world here- it's just audio.

Keep it civil, fun and enjoyable, sprinkle in facts, measurements and a bit of discourse. Not too much. I'd want to have a drink with the people I interact with here, not end up in a bar fight and get thrown out. :)
 

kotmj

Member
Joined
Jun 14, 2021
Messages
38
Likes
13
I've asked removing my username already twice so just do it and don't write any threats.
Yes, just keep it cool. Just remind yourself that humans are inherently limited, and accept their limitations.
 

Ageve

Member
Joined
Jun 13, 2021
Messages
16
Likes
35
Not sure if you can call it a direct replacement. Here the spektor 1 costs ~38% lower and has a 4" woofer instead of the Zensor 1's 5".

Spektor 2 may be the direct replacement for Zensor 1, but it's ~40% more expensive (at least here in Sweden), and the build quality is not as good.

Spektor 1 costs the same as Zensor 1 did here (~230 USD / pair).
 
Top Bottom