I just had a read of the Smith paper and didn't find much relevant to the current discussion.
Firstly,
@levimax, the definition of transient response you're working with is not the definition used in that paper. The paper is interested in the classical definition of transient response, which encompasses both ringing and group delay. Given that the paper investigates only relatively low-order filters (1st and 2nd order), ringing is negligible, and it is group delay that is responsible for the observed differences between the filters used. I don't think there's enough info in that paper really to draw any conclusions, as the individual responses of the filters/drivers used are only described in broad terms, and were not measured.
Secondly, the filter types used in the passive networks were then-standard Butterworth filters, which have now fallen out of favour due to their inability to correctly sum at the crossover point in the frequency domain. In most modern passive speakers, these Butterworth networks have been replaced by Linkwitz-Riley networks or variations thereof, which do sum correctly in the frequency domain. The active filters used in the study were Crowhurst filters, which are a filter type now long-forgotten but which (from what I can gather from the online sources available) were optimised for transient response and which were technically superior to the Butterworth networks to which they were being compared.
So it seems to me that all the paper demonstrates is that by using active filters, the authors were able to achieve better transient response than with the Buterworth passive networks originally used in the speakers they investigated. This did of course result in better transient response, but since the authors did not make any effort to design their own passive networks, all they are really comparing are the speakers' stock passive networks with the authors' customised active networks.
From this I think the only conclusions that can be drawn are that:
- the stock passive networks were sub-optimal
- it was easier to produce a speaker with superior transient response using the active filter networks chosen by the authors
These conclusions are both correct but, at the end of the day, the authors were comparing now-outdated passive filter topolgies with a now-outdated active filter toplogy.
Things have obviously moved forward a long way since that paper, and of course modern DSP-based active filters are capable of far better transient performance than both the passive and active networks used in the paper.
Finally, the authors began from the assumption that the degree of group delay introduced by the passive networks they looked at was audible. Even half a century later,
this assumption is yet to be experimentally demonstrated.
There is some more information on the Crowhurst networks the authors investigated
here.
In any case, I think that paper is of little interest to the present discussion.