Please read FAQ before…

c Expand All C Collapse All

Measurements

The method is very stable. To check it, just do some successive measurements and compare. We encourage to also measure at different levels to check consistency of your setup and your corrections.

slovenska-lekaren.com

In case you see differences in low frequencies, this may come from some conditions that have varied between measurements : are doors and windows allways in same positions, opened or closed ?

Question frequently arises : how large should you scan around listener head position ? Move in all directions, even height, but avoid to be too near  of a reflecting surface, ie stay a bit higher than the back of the chair.

I would scan within about +-15° off axis, that is nearly 50cm in all directions for a loudspeaker at 2m and 100cm at 4m.

Remember that for a home audio system, the MMM scan is mainly to smooth out reflections and not to average for more listeners.

When you adjust the sound level so that the voice is nearly same as a speaking women in your room, the level of sweeps is between 75 to 80dB SPL(C), pink noise on one channel is also about 75 to 80dB(C) and with both channels together between 80 to 85dB(C). This gives you an indication at which level distortion is measured.

For many people, most curves and graphs are not so easy to understand so some of you have asked about a simple performance rating. We have tested the ratings proposed by Sean Olive in AES papers 6113 and 6190 but for some reasons, it was not totally satisfying. Those ratings are based on anechoic room measurements extended to Predicted In Room results. With our method, we only measure InRoom values and we have to quantify performance based only on those real measurements.

We get the score from three main factors :

  • SM_IRR SMoothness of InRoom Response between 125 and 11500Hz : the proposal of Olive is not very intuitive (Pearson coefficient) and this value is not used by us
  • NBD Narrow Band Deviation of InRoom Response between 125 and 11500Hz (6.5 octaves) : measured surface difference between 1/20th octave curve and 1/2 octave curve, so it is not related to target and general slope
  • WBD Wide Bandwidth Deviation of frequency response from target curve : it is a value based on area difference (so related to variance) between the measured response and the target response between 125 and 11500Hz
  • LFD Low Frequencies Deviation is based on area difference between the measured response and the target response between 25 and 125Hz (2 octaves) but calculated on a linear frequency scale. We use a linear scale because we consider that problems at the high part of this frequency range are more audible and problematic than at the vey low frequencies. Note that this low frequencies target is flat under 80Hz, so it not not exactly same as the LF target defined in the Upload form
  • please notice that the displayed mean value is the lowest of L and R values or L, C and R in case of multichannel

Global performance rating = 0.25*NBD + 0.4*WBD + 0.35*LFD
It is important to understand that the rating is only based on measured amplitude response and is missing other factors that may influence audible quality : max levels, directivity, distortions, phase and time response, etc… So be carefull when you compare ratings of different systems, ie the highest may not be the best ! But compare numbers before/after equalisation/correction is certainly valid.

 

If field “Measure and correct” is validated, subdirectory “Correction” is created to contain.wav files for FIR correction, respectively Left and Right in linear phase and minimal phase : xxx-hyblinL.wav, xxx-hyblinR.wav, xxx-hybminL.wav, xxx-hybminR.wav. Those files can be directly used for corrections.

Pages p7 to p9 are also created.

p7 Correction
Separated C1 measurements L and R and also L+R (C2 black)
FIR corrections C3 calculated from L, R and L+R and hybrid corrections C4 (same in lower frequencies and separated above)
C5 Phase correction
Simulated responses of L, R and L+R after FIR correction : C7 for separated corrections and C8 for hybrid correction
p8 Simulated ETC energy-time curve after correction for pre-echo visualisation
p9 Simulated wavelets for pre-echo visualisation

Exponential sine sweeps (promoted by A. Farina) is a very nice technique to measure loudspeakers response with a good rejection of noises. But it may have problems when the player or recorder clock are unlocked (ie when you record on a computer that is independant of the player). In this case, the displayed impulse and step responses, spectrograms, wavelets,… may be distorded (don’t worry, the MMM part and so frequency response and correction are not affected). To avoid those problems, we have added in LA2v3 version, another time sync signal at the end of the audio sequence to automatically compensate for clock mismatch. The value of drifted samples is indicated in page p0. If mismatch is less than 2000ppm (CD Redbook specifications is 200ppm), the recording is resampled. This corresponds to about +-15360 samples for the stereo file. If mismatch is higher than 15360ppm, this is not considered as a clock difference but as a recording problem (ie usb dropouts,…) and so the recording is not resampled.

If some folders are empty without any graphs, it may be that the user did not upload any measurement, or deleted the files or uploaded measurements as private.