Vanatoo Transparent Zero Speaker Review

[Sancus]

Really, EQ is a linear change so unless we are dealing with compression and such, its effect should be similar to simulation.

If this is the case, and these simulations are pretty close to reality, and you can achieve a large improvement in a short time with a small number of adjustments, why wouldn't the designer of the speakers in question make the changes? Talking purely about actives here, of course. Though even with passives, it seems like the manufacturer could include information to help you EQ.

I mean, the scores above are arguably as good or better than a Genelec 8341A, starting with a considerably worse baseline in terms of hardware. If it's really that simple, why isn't everyone doing it? Or is the preference score really not reflecting reality when used this way?

 

[QMuse]

Really, EQ is a linear change so unless we are dealing with compression and such, its effect should be similar to simulation.

Exactly.

 

[QMuse]

Or is the preference score really not reflecting reality when used this way?

Even if preference score isn't reflecting reality EQ is based on Klippel's spinorama measurements which are reflecting reality. And that means if EQ makes LW and PIR smoother speaker will sound better.

 

[Sancus]

Even if preference score isn't reflecting reality EQ is based on Klippel's spinorama measurements which are reflecting reality. And that means if EQ makes LW and PIR smoother speaker will sound better.

Yes. My fundamental question is that if this is such simple, low hanging fruit, that you can make a $150 7" tall speaker(with a sub, granted) sound better than a $3000 Genelec, why isn't anybody, even JBL/Harman, just building such EQ into their speakers? I mean we've seen expensive actives like the 705P reviewed, and they're clearly not doing it. They have access to anechoic measurements and the NFS itself, so that's not the issue.

What gives?

 

[HooStat]

Really, EQ is a linear change

But I don't think the score is linear. Or, at least, it shouldn't be. Depends on how they modeled it. And that might be the problem -- if they used a linear regression model then the scores near the upper bound (i.e., 9-10) are likely to be estimated inaccurately. The model probably has no problem estimating an 11 or 12 even thought that is not possible given the ratings (which seem to be 0-10). It would be interesting to max out all the covariates in the model and see what it predicts. Does anyone have the actual model? Sorry if it has been posted somewhere. It would be interesting to look at.

 

[QMuse]

Yes. My fundamental question is that if this is such simple, low hanging fruit, that you can make a $150 7" tall speaker(with a sub, granted) sound better than a $3000 Genelec, why isn't anybody, even JBL/Harman, just building such EQ into their speakers? I mean we've seen expensive actives like the 705P reviewed, and they're clearly not doing it. They have access to anechoic measurements and the NFS itself, so that's not the issue.

What gives?

You can't make $150 speaker to sound better than a $3000 Genelec with EQ but you can make it sound better than original, in some cases much better. The same goes for room EQ part and yet more than 90% of audiophiles are not doing it.

I can also imagine that if manufacturer publishes recommended speaker EQ based on their NFS measurements it would imply big obligation on their part to keep inter-sample variations small.

 

[QMuse]

But I don't think the score is linear. Or, at least, it shouldn't be. Depends on how they modeled it. And that might be the problem -- if they used a linear regression model then the scores near the upper bound (i.e., 9-10) are likely to be estimated inaccurately. The model probably has no problem estimating an 11 or 12 even thought that is not possible given the ratings (which seem to be 0-10). It would be interesting to max out all the covariates in the model and see what it predicts. Does anyone have the actual model? Sorry if it has been posted somewhere. It would be interesting to look at.

That's not what he meant. What he wanted to say is that if you apply EQ filter to boost (say) 1kHz by 3dB with Q=3, the speaker will react in linear proportion to it and will provide pretty much the same output.

 

[MZKM]

But I don't think the score is linear. Or, at least, it shouldn't be. Depends on how they modeled it. And that might be the problem -- if they used a linear regression model then the scores near the upper bound (i.e., 9-10) are likely to be estimated inaccurately. The model probably has no problem estimating an 11 or 12 even thought that is not possible given the ratings (which seem to be 0-10). It would be interesting to max out all the covariates in the model and see what it predicts. Does anyone have the actual model? Sorry if it has been posted somewhere. It would be interesting to look at.

You can’t get above a 10, but you can go below 0.

 

[Sancus]

You can't make $150 speaker to sound better than a $3000 Genelec with EQ but you can make it sound better than original, in some cases much better. The same goes for room EQ part and yet more than 90% of audiophiles are not doing it.

The Vanatoo scores 9.2 w/EQ and a sub, 0.79 better than the 8341A. That's enough to be preferred over the Genelec with >50% probability and certainly enough to be competitive. Assuming the preference score is still valid when manipulated with EQ and simulated this way. Which is an additional assumption, to be sure.

 

[QMuse]

The Vanatoo scores 9.2 w/EQ and a sub, 0.79 better than the 8341A. That's enough to be preferred over the Genelec with >50% probability and certainly enough to be competitive. Assuming the preference score is still valid when manipulated with EQ and simulated this way. Which is an additional assumption, to be sure.

I see no reason why preference score wouldn't be valid after EQ as it is based on the same curves that EQ affected. I am not however sure in the relation between preference score and subjective score as @amirm rating doesn't really correlate well with preference rating.

 

[HooStat]

You can’t get above a 10, but you can go below 0.

That doesn't make sense unless they calibrated it to max out at 10. Which means they are simply masking the non-linearity. Can someone point me to the source for this model? I would like to see the model and the coefficients.

 

[BYRTT]

If this is the case, and these simulations are pretty close to reality, and you can achieve a large improvement in a short time with a small number of adjustments, why wouldn't the designer of the speakers in question make the changes? Talking purely about actives here, of course. Though even with passives, it seems like the manufacturer could include information to help you EQ.

I mean, the scores above are arguably as good or better than a Genelec 8341A, starting with a considerably worse baseline in terms of hardware. If it's really that simple, why isn't everyone doing it? Or is the preference score really not reflecting reality when used this way?

Yes. My fundamental question is that if this is such simple, low hanging fruit, that you can make a $150 7" tall speaker(with a sub, granted) sound better than a $3000 Genelec, why isn't anybody, even JBL/Harman, just building such EQ into their speakers? I mean we've seen expensive actives like the 705P reviewed, and they're clearly not doing it. They have access to anechoic measurements and the NFS itself, so that's not the issue.

What gives?

Cant know for sure but think some reason in real world scenarios can be lack of DSP resources in imagine those precise reverse EQ curves used here would demand probably few hundreds of real PEQ adjustments to produce the same detailed EQ correction curve, that's one reason myself love to use Jriver as DSP to steer a active speaker system because there is no limit on numbers of corrections other than the real time CPU horsepower level should match up, think anechoic raw that Genelec looks beatifull and scored 6,71 and if it happen @MZKM is alright more work i attach EQ filter below to correct 8341A to same policy for a smooth listeneng window that is covering +/-30 deg horizontals and +/- 10 deg verticals at 2 meter according to standard of CTA-2034.

 

[HooStat]

That doesn't make sense unless they calibrated it to max out at 10. Which means they are simply masking the non-linearity. Can someone point me to the source for this model? I would like to see the model and the coefficients.

Just to be clear, I am not saying this is "wrong". I am just trying to understand how to best interpret the model.

 

[MZKM]

That doesn't make sense unless they calibrated it to max out at 10. Which means they are simply masking the non-linearity. Can someone point me to the source for this model? I would like to see the model and the coefficients.

People were asked to rank the speakers on a scale from 0-10, and the formula was made to best predict how other speakers would rank, it should be common sense that the best outcome could be a 10.

 

[txbdan]

What is this Preference Rating stuff? I don't see any stickies or "formal" info on what it is, who's doing it, with what info, etc.

Edit: nevermind, i found it in the General Audio subforum.

 

[HooStat]

People were asked to rank the speakers on a scale from 0-10, and the formula was made to best predict how other speakers would rank, it should be common sense that the best outcome could be a 10.

That is true. But the statistical model doesn't know that, particularly if they used linear regression. If so, the model itself has no problem predicting any continuous value including values > 10. That is why I was asking if anyone had ever put in maximum values for all inputs (or simply really good values) and tested to see what the predicted value was. I suspect that you can get some pretty high preference scores from the model.

 

[MZKM]

That is true. But the statistical model doesn't know that, particularly if they used linear regression. If so, the model itself has no problem predicting any continuous value including values > 10. That is why I was asking if anyone had ever put in maximum values for all inputs (or simply really good values) and tested to see what the predicted value was. I suspect that you can get some pretty high preference scores from the model.

The LFX (Low Frequency Extension) could tilt it past 10, but it would need extremely deep bass, not even the Rythmik subwoofer measured would be enough, while also being near perfect in the other categories. So yes, it’s theoretically possible (it can even go past 100), but not even remotely practical.

 

[amirm]

I see no reason why preference score wouldn't be valid after EQ as it is based on the same curves that EQ affected. I am not however sure in the relation between preference score and subjective score as @amirm rating doesn't really correlate well with preference rating.

But I don't think the score is linear.

I am talking about the input to the preference model, not output. A linear change to the frequency response can simply be computed has has been in these threads. The only thing it can't predict is if we have a lot of distortion and that bleeds into frequency reponse.

 

[HooStat]

The LFX (Low Frequency Extension) could tilt it past 10, but it would need extremely deep bass, not even the Rythmik subwoofer measured would be enough, while also being near perfect in the other categories. So yes, it’s theoretically possible (it can even go past 100), but not even remotely practical.

Just out of curiosity, do you ever provide the model inputs? Or just the model outputs only. How many covariates (inputs) are there?

 

[MZKM]

Just out of curiosity, do you ever provide the model inputs? Or just the model outputs only. How many covariates (inputs) are there?

What do you mean? You mean the value each has after individual computation and what gets put into the final formula? If so, no, but I do show that radar chart which turns it into a 0-100% scale for each variable (using what values obtain a 0 or 10).

If you want to know the approximate values for the Vanatoo:

NBD_ON: 0.41
NBD_PIR: 0.44
SM_PIR: 1.78
LFX: 0.52

Rating: 12.69 - 2.49(NBD_ON) - 2.99(NBD_PIR) - 4.31(LFX) + 2.32(SM_PIR)

https://www.audiosciencereview.com/...quivalent-sinad-discussion.10818/#post-302047