Sony SS-CS5 3-way Speaker Review

[andreasmaaan]

This was actually done in the study. The regression analysis was performed by testing the same budget bookshelf speakers consumer reports had tested. Once developed, they then applied it to larger set of speakers. That is when the correlation factor dropped from 0.99 to 0.86. It was then hypothesized that there were larger speakers in there that were not modelled as well as the smaller ones.

This is not quite right IIUC.

In Olive's second study (involving the larger sample set), he states the following:

Our anechoic model described in equation 9 [i.e. the model based on the first, smaller sample set] was applied to the new larger loudspeaker sample and produced a correlation of 0.70 between the predicted and measured preference ratings. The lower correlation was likely related to the model being too tightly fitted to the small sample (13 loudspeakers) and/or the loss of precision from combining subjective data from 18 unrelated tests. A more generalized model was necessary to accurately predict the ratings of our 70-loudspeaker sample. Using 23 independent variables, a model using 4 independent variables was developed that has a correlation of 0.86 for the 70-loudspeaker sample.

In other words, the model derived from the first (smaller) sample set produced a correlation factor of 0.70 when applied to the second (larger) sample set.

A new model was then derived from the second sample set. It was this new model that produced a correlation factor of 0.86 when applied to the same sample set from which it was derived.

 

[Sancus]

A new model was then derived from the second sample set. It was this new model that produced a correlation factor of 0.86 when applied to the same sample set from which it was derived.

Probably correct, but doesn't Olive basically say this at the end when he says a possible limitation is that "The model is limited to the specific types of
loudspeakers in our sample of 70."

I think this is a less interesting criticism than it seems off-hand though as most of the speakers reviewed so far on ASR are very conventional types, with a few exceptions. I do think it's entirely possible that more exotic designs like Genelecs, D&D 8C, or Devialet Phantoms would be even more overrated or underrated(no way to be sure) than the model predicts.

It's a bit unfortunate that we don't know exactly what types at what performance levels were included in the study, as well.

 

[andreasmaaan]

Probably correct, but doesn't Olive basically say this at the end when he says a possible limitation is that "The model is limited to the specific types of loudspeakers in our sample of 70."

Absolutely. I'm not meaning to imply dishonesty on Olive's part. In fact, I'm simply trying to emphasise here what he says about his own work.

I think this is a less interesting criticism than it seems off-hand though as most of the speakers reviewed so far on ASR are very conventional types, with a few exceptions. I do think it's entirely possible that more exotic designs like Genelecs, D&D 8C, or Devialet Phantoms would be even more overrated or underrated(no way to be sure) than the model predicts.

It's a bit unfortunate that we don't know exactly what types at what performance levels were included in the study, as well.

I think my point is an important one, as it adds a major, often not recognised/understood caveat to statements about this model's predictive power.

I agree that lack of information about the speakers in the sample is an additional/compounding problem.

 

[BenB]

This is not quite right IIUC.

In Olive's second study (involving the larger sample set), he states the following:



In other words, the model derived from the first (smaller) sample set produced a correlation factor of 0.70 when applied to the second (larger) sample set.

A new model was then derived from the second sample set. It was this new model that produced a correlation factor of 0.86 when applied to the same sample set from which it was derived.

Thanks for the information. I was never able to read the studies before, and all my information came from Toole's book, and whatever was divulged on AVS. I did request a copy of the study from Olive, but never heard from him. Thanks to your quote, I was able to find part II of the study online. I'm trying to see whether or not the original 13 loudspeakers are included in the larger set of 70. If they are, then even the 0.70 number is exaggerated as a measure of the original algorithm's predictive abilities.

 

[andreasmaaan]

Thanks for the information. I was never able to read the studies before, and all my information came from Toole's book, and whatever was divulged on AVS. I did request a copy of the study from Olive, but never heard from him. Thanks to your quote, I was able to find part II of the study online. I'm trying to see whether or not the original 13 loudspeakers are included in the larger set of 70. If they are, then even the 0.70 number is exaggerated as a measure of the original algorithm's predictive abilities.

IIRC, the second sample doesn't include the 13 from the original sample, but don't quote me

 

[MZKM]

exceptions. I do think it's entirely possible that more exotic designs like Genelecs, D&D 8C, or Devialet Phantoms would be even more overrated or underrated(no way to be sure) than the model predicts.

The formula very much expects a typical slope, so even very wide dispersion monopole speakers (like Dennis‘ Philharmonic BMR) would likely get a poor score, let alone say an omnidirectional speaker.

I would love for the data to be made public so there could be experimentation, but that is likely a pipe dream.

I am doing some experimentation with seeing if I can create a modified score where the PIR curve is made flat (so the slope doesn’t matter), but I don’t have much luck yet, as there are many coefficients that would need to be tinkered with and one would need to do this on a decent amount of speakers with what the paper says is ideal slope to see if a universal modification can be created.

 

[Wombat]

What the heck???

View attachment 65195

Amazon shows single port as I have.

I think marketing provided the wrong photo.

 

[amirm]

I think this is a less interesting criticism than it seems off-hand though as most of the speakers reviewed so far on ASR are very conventional types, with a few exceptions.

I have a very oddball speaker to test in the future.

 

[amirm]

I think marketing provided the wrong photo. View attachment 65276

Or it had two but due to the need for social distancing, they moved one to the other speaker....

 

[richard12511]

For those with expertise in statistics, should it strike me as problematic that these correlation percentages were determined with reference to the same sample sets from which the model was derived?

I don't know, but I'm starting to think yes. I'm really starting to doubt that those .86 and .99 correlation figures hold up in today's market.

I'd love to hear your thoughts, though. You know way more about this than I do.

 

[napilopez]

The formula very much expects a typical slope, so even very wide dispersion monopole speakers (like Dennis‘ Philharmonic BMR) would likely get a poor score, let alone say an omnidirectional speaker.

I would love for the data to be made public so there could be experimentation, but that is likely a pipe dream.

I am doing some experimentation with seeing if I can create a modified score where the PIR curve is made flat (so the slope doesn’t matter), but I don’t have much luck yet, as there are many coefficients that would need to be tinkered with and one would need to do this on a decent amount of speakers with what the paper says is ideal slope to see if a universal modification can be created.

Amir and I talked about it a little in another thread, but it might be interesting to investigate Paper 1 for a bit deeper, as that paper:

• Has a higher accuracy score
• Involves only bookshelves/non full range speakers, which is most of what Amir is testing so far
• Includes extensive subjective listening comments/impressions for all 13 speakers.
• Crucially, includes a spinorama for all 13 speakers.

The spinoramas should be fairly easy to digitize from the paper, and we could probably even identify each of the speakers with a little bit of investigation as they were all top-ranked by Consumer reports. If someone is up for it, it could be interesting to see if the correlations with Amir's impressions change vs the generalized model.

I do think it's entirely possible that more exotic designs like Genelecs, D&D 8C, or Devialet Phantoms would be even more overrated or underrated(no way to be sure) than the model predicts.

I agree with your premise, though the 8c and Reactor seem in line with what I'd expect from the formula based on my own messing with MZKM's sheet. No big surprises with the speakers I've plugged in yet.

Disclaimer: there's obviously way more variability and fudging with computing the scores based on my quasi-anechoic measurements. Like, bass extension can easily be modified depending on where I choose to splice, etc, and I obviously have less resolution at lower frequencies. It means I miss out on potential high q resonances, though to my surprise, having lower resolution seems to actually hurt most scores rather than help it based on my tests so far.

Anyway, the 8c gets a 7.4/8.4 based on my measurements, for instance. The Reactor gets a 7.2/7.6. those both seem pretty much in line with what I'd expect from my quasi-anechoic spins. They both get super high no-sub scores because they both have ridiculous bass extension at the SPL level tested. The Reactor suffers because though the overall curves are smooth and flattish, the on-axis is jagged.

I don't know if we'll ever seen a speaker cross a 9 lol. But yes it'd be interesting to see how an omni or super wide speaker does. I'm going to try to see what I get for the Sonos move, which is the closest I have to that, but I expect it wont do too great due to the jagged on-axis.

 

[tomtoo]

@QMuse
Great! Looks you nailed @armirs taste!

 

[tomtoo]

This is realy a great thread! And i'am so happy that @amirm decidet to give his subjective impressions too.

 

[andreasmaaan]

BTW @

I don't know, but I'm starting to think yes. I'm really starting to doubt that those .86 and .99 correlation figures hold up in today's market.

I'd love to hear your thoughts, though. You know way more about this than I do.

The thing is, unfortunately I don't know statistics. I'm enrolled to study it later this year though - maybe ask me in 6 months

But in terms of my intuitive/commonsense view (and being quite aware that "common sense" is often highly misleading), I would imagine that the probability that a model derived from sample set S correlates with sample set S is higher than it is for all other sample sets, and in particular for a larger sample set that includes S (i.e. the set of all loudspeakers) - unless S is perfectly representative, of course.

It's basically this proposition that I'd like someone with a stats background to verify/refute.

 

[QMuse]

I just plugged this into the convolution filter in Roon since it was easier than programming the PEQ for the other proposal.

Wow! What a transformation. Detail is there. Tonality is warm. High frequency harshness is gone mostly. Fidelity is back in a big way! Not perfect mind you but boy, is the speaker transformed.

I would say it is now a steal for $78 or even more if someone uses this filter or something like it.

Note that it did accentuate the room mode, making it sound boomy so I had to dial in my 100 Hz filter back in.

I'm so glad it worked!

Btw, filter didn't alter freqs below 400Hz but the overall tonal balance has changed in favor of bass so your room EQ filter is definitely necessary.

 

[QMuse]

Very interesting how Amir greatly preferred the EQ done by Qmuse. This one had more perceived bass, so maybe that's it. We of course don't know how these filters effect THD and compression, so that's a factor.

The scores are close enough that the inherent error of the formula can be used such that one shouldn't look too closely into this.

Here's how my filter looks. As you can see it is very conservative, all PEQs are with Q<=3 and there is no boost except only at the end of the range, which is more for the curve to look smooth than to have audible effect. The idea of this filter was to get smooth tonal balance which I thought @amirm would prefer.

 

[QMuse]

The formula very much expects a typical slope, so even very wide dispersion monopole speakers (like Dennis‘ Philharmonic BMR) would likely get a poor score, let alone say an omnidirectional speaker.

I would love for the data to be made public so there could be experimentation, but that is likely a pipe dream.

I am doing some experimentation with seeing if I can create a modified score where the PIR curve is made flat (so the slope doesn’t matter), but I don’t have much luck yet, as there are many coefficients that would need to be tinkered with and one would need to do this on a decent amount of speakers with what the paper says is ideal slope to see if a universal modification can be created.

I think PIR slope should matter, in other words I think most of us prefer certain slope over the others.

 

[bigx5murf]

I still want to see the Sony SS-HA1 measured.

 

[GXAlan]

I still want to see the Sony SS-HA1 measured.

I had a SS-HW1 which I recently sold. It’s a much higher end product than the CS5 so the exaggerated treble that Amir reported wasn’t seen. In room, it sounded better than Tag McLaren Calliopes although the Calliopes has better resolution, clarity, and speed. It was just that the SS-HW1 has a warmer midrange and I thought better in room dispersion.

It is 83 dB/2.83V so you know it isn’t designed for mass market. The SS-HW1 is a similar appearing speaker to the SS-HA1 but has a better woofer and surprisingly is one of the speakers supported by Devialet SAM.

All that said, the SS-HW1 did not sound as good as the JBL XPL-90 which was measured here.

All sighted bias.

Sony makes exotic speakers such as this one (use google translate)
https://www.hifido.co.jp/sold/20-25953-02333-00.html

 

[Dmi3i]

Excellent review, thanks Amirm. Has anyone listened to the Q acoustics 3050i speaker system? I found an objective test of the 3050i https://www.soundstagenetwork.com/i...&catid=77:loudspeaker-measurements&Itemid=153. There are questions around 250 Hz, but for the money I think this is a great option, and the distortion is suspiciously low ... what do you think?