Hello everyone,
I'm interesting in the problem of finding a target curve for a stereophonic system, i.e. a system which creates the illusion, for left and right coherent signals (same amplitude and same phase), that a physical source is present in front of the listener. However, this trick implies some coloration of the signal due to HRTF.
In the case of a free field, HRIR database exists, measured in an anechoic for different people.
I've used some of them (ITA database with 13 individuals) to calculate the impulse response of a L/R coherent signal. I've done the following :
- calculation of the filter applied to the L/R coherent signals for left ear and then for right ear for each individual,
- time alignment and then vector averaging of this filter for 13 individuals
- assuming symmetry for both individual anatomy and source placement, I time aligned and vector averaged the resulting filters to get a new filter, let's call it SX, where X is the angle of the source from the center from the listener point of view.
I've done it for the following degrees : {0, 5, 10, 15, 20, 25, 30, 35, 40, 45}
Then, I calculated the impulse response S0 / SX for X in {5,10, ..., 40,45} to get the difference in SPL from the listener point of view.
Because some people may argue that a stereo recording is mastered to be played on stereo, and thus the reference impulse should not be the one located in front of you but the one on a stereo configuration with speakers placed at -30/+30°, I also did the calculation of S30 / SX for X in {5,10, ..., 25, 35, ..., 40, 45}
Here is the database HRTF I used.
Here is the Matlab script I used to extract the 4 relevant HRIR, (left ear left source, left ear right source, right ear left source and right ear right source), from the 2304 available for each individuals (comments are in French sorry, I can translate if some people want it), adding the left ear left source + left ear right source in frequency domain (obvisously I did the same for the right ear) then export the resulting impulse in a .wav file to be imported in REW for time alignment and vector averaging.
Here are all the calculated impulses in .mdat format.
I obtained the following curves, with 1/2 octave smoothing :
Using front mono source as impulse reference :
Using 30° stereo configuration as impulse reference :
What do you think about this approach ? what do you think about direct field / diffuse field importance in perceived sound ? My room is very dead (0.170ms for 100m^3), I never liked reverberation because of some peaking frequency always existing that bleed my ears, but I like reverberation for the natural sounding too, and the ability to mask "defaults" in some recordings that sound dull otherwise. Do you have impulse measurements to share (in .mdat format) so that I can give a look to your spectrogram ? My next step will be to fake a diffuse field with my speakers.
Thank you for your
I'm interesting in the problem of finding a target curve for a stereophonic system, i.e. a system which creates the illusion, for left and right coherent signals (same amplitude and same phase), that a physical source is present in front of the listener. However, this trick implies some coloration of the signal due to HRTF.
In the case of a free field, HRIR database exists, measured in an anechoic for different people.
I've used some of them (ITA database with 13 individuals) to calculate the impulse response of a L/R coherent signal. I've done the following :
- calculation of the filter applied to the L/R coherent signals for left ear and then for right ear for each individual,
- time alignment and then vector averaging of this filter for 13 individuals
- assuming symmetry for both individual anatomy and source placement, I time aligned and vector averaged the resulting filters to get a new filter, let's call it SX, where X is the angle of the source from the center from the listener point of view.
I've done it for the following degrees : {0, 5, 10, 15, 20, 25, 30, 35, 40, 45}
Then, I calculated the impulse response S0 / SX for X in {5,10, ..., 40,45} to get the difference in SPL from the listener point of view.
Because some people may argue that a stereo recording is mastered to be played on stereo, and thus the reference impulse should not be the one located in front of you but the one on a stereo configuration with speakers placed at -30/+30°, I also did the calculation of S30 / SX for X in {5,10, ..., 25, 35, ..., 40, 45}
Here is the database HRTF I used.
Here is the Matlab script I used to extract the 4 relevant HRIR, (left ear left source, left ear right source, right ear left source and right ear right source), from the 2304 available for each individuals (comments are in French sorry, I can translate if some people want it), adding the left ear left source + left ear right source in frequency domain (obvisously I did the same for the right ear) then export the resulting impulse in a .wav file to be imported in REW for time alignment and vector averaging.
Here are all the calculated impulses in .mdat format.
I obtained the following curves, with 1/2 octave smoothing :
Using front mono source as impulse reference :
Using 30° stereo configuration as impulse reference :
What do you think about this approach ? what do you think about direct field / diffuse field importance in perceived sound ? My room is very dead (0.170ms for 100m^3), I never liked reverberation because of some peaking frequency always existing that bleed my ears, but I like reverberation for the natural sounding too, and the ability to mask "defaults" in some recordings that sound dull otherwise. Do you have impulse measurements to share (in .mdat format) so that I can give a look to your spectrogram ? My next step will be to fake a diffuse field with my speakers.
Thank you for your