Hello guys,
We are just on the beginning of something great, so we rather carefully adapt and further perfect the procedure now before we regret it later after having measured dozens of loudspeakers.
I agree that things should not be rushed.
About the measurement level, here are the results of old measurements that I made at home with REW, that uses Farina deconvolution to measure the frequency response (measurements with a sine sweep, I guess that the Klippel uses the same method), at various levels.
I made 4 identical measurements in a row, the playback volume being the only thing that changes between the curves. The vertical position of the curve was not modified, and shows the actual volume difference.
I didn't measure the real SPL level. I guess that the maximum level was above 80 dB at 2 meters, with both speakers playing. The background noise in the room is around 35 dB SPL(A).
Fort = Loud / Faible = Quiet
They look similar. The Farina method has a weak sensitivity to background noise. I've asked REW to show the difference between these curves (arithmetic division) :
Here, it is very interesting to see that the difference curves at -10 and -20 look the same. Which means that the odd man out would be the loudest measurement, here taken as a reference.
If we trace the difference between the two intermediate curves, here is what we get :
I Don't think the speaker were playing at more than 90 dB at 1 meter. I am too afraid to fry a tweeter by pushing the volume during measurements.
In spite of the moderate level, compared to the >100 dB at 1 meter, we can see that the playback level is a stronger source of inaccuracy than the background noise of the room… at least during this measurement.
Therefore I agree that some more experiments would be interesting in order to choose the right playback level for measurements.
However, my opinion about the reference axis is different. I think that all speakers should be measured with the tweeter as a reference axis, even if the manufacturer says otherwise.
The two reasons are :
The listening window, early reflections, and power curves can only be compared to the tweeter curve. Comparing them to another reference curve would introduce a lot of confusion about the directivity of the speaker. The listening window curve might even be louder than the custom reference on-axis curve is some cases, if that one is not taken in front of the speaker ! How could we interpret that ? Every speaker measurement would have to be interpreted differently, since the spinorama curves would be measured differently. I Wonder if we might not be breaking the CEA 2034 spec if we followed the manufacturer recommendations for the reference axis !
And secondly, when we are listening to speakers in a room, the room doesn't care if the reference axis is in front of the tweeter or not. The predicted in-room response, (as well as the actual in room response, by the way), will be the result of the speakers complete acoustic behaviour. We must expect a speaker calibrated for flat response at mid-height measurement to sound brighter than a speaker calibrated for flat response at tweeter's height… and we want to see this difference on Amir's measurements, since that will make the speakers sound actually different !