Hi all,
I'd like to analyse recordings to see if low frequencies are correlated (most of recordings are) or not, especially within reverberation, and have kind of a "room spatiousness" indication.
For this, I can do a comparison between L+R and L-R but if some low frequency sources/instruments are more or less central, this comparison gives worthless indication.
So I thought of comparing histograms of lowpassed L+R and L-R and look at the low amplitude range.
Hereunder are histograms of some tracks with low frequency content :
Bird on a wire (Jennifer Warnes) and Fast Car (Tracy Chapman) : low frequency is mostly mono and lowpassed (150Hz) L+R is higher than L-R
Billy The Kid and Pentecost Mass (recorded by John Eargle for Delos) are AB recordings : lowpassed L+R is lower or equal than L-R
(https://i62.servimg.com/u/f62/11/21/77/49/511.png)
Do you think that this method is valid or do you think of a better one ?
I'd like to analyse recordings to see if low frequencies are correlated (most of recordings are) or not, especially within reverberation, and have kind of a "room spatiousness" indication.
I don't understand what you're trying to do, but it's "well known" that bass is non-directional. And it makes sense when you think about the wavelengths relative to the space between your left & right ears. But, I don't know at what frequency sound starts to become directional.
Are you doing any listening tests?
It's not going to be easy to do a directional-sensitivity experiment at home because of the acoustics... Moving a speaker or panning will likely affect the amplitude so it's likely you'll hear a difference and it might be hard to tell if you're
really hearing a directional-difference. Any distortion or rattling, might also give-away the direction.
I'm surprised there's so much L-R information. Whenever I've subtracted the channels (or reversed the polarity of one speaker) the bass "goes away". But, my perception of "bass" is probably below 150Hz, so you may be getting a difference signal at the upper end of your pass band.
Even if bass was directional, you'd generally want it mixed as mono to utilize the piston area of both woofers and the power of both amplifier channels. And, it would be silly to be pumping-out out-of-phase bass and wasting amplifier power/headroom on sound that's going to be acoustically canceled.
Thanks Doug, but it is not about acoustic measurement, it is to try to measure how reverberation is decorrelated in recordings, measuring the content of the recorded wav file.
Please have a look at Dave Griesinger's papers ie here (http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.41.8575&rep=rep1&type=pdf). If you directly measure L-R, you generally find quite nothing. I try to find a way to measure L-R of the reverb only.
Bird on a wire (Jennifer Warnes) and Fast Car (Tracy Chapman) : low frequency is mostly mono
Try classical and jazz. There are AES papers that I've linked previously, about stereo LF content in recordings.
it's "well known" that bass is non-directional.
Maybe, but still wrong.
Bird on a wire (Jennifer Warnes) and Fast Car (Tracy Chapman) : low frequency is mostly mono
Try classical and jazz. There are AES papers that I've linked previously, about stereo LF content in recordings.
I used those two tracks because they have a nice low frequency content but mostly mono so to compare with the two others (Billy and Pentecost) that also have low frequency content but less correlated.
Hi, I did some more tests, here is another way to show correlation.
Instead of analysing the histograms, I think it is better to analyse the temporal structure directly : we need to check correlation preferably in the reverberation parts, where it is audible !
Ie if you record classical music in large AB setup, cellos maybe quite central and correlated and when cellos are playing, you will see no decorrelation. But when music stops and only reverb stays, decorrelation will show up.
Same for pop/rock/..., standard pan-potted stereo will stay correlated, it is only in the reverb tail, you may see some decorrelation added by the electronic reverb.
Same tracks as above, (x-axs is time in seconds), peak and rms values are shown by color lines, correlation is black line : the -60dB line separates uncorrelation (above) and correlation (under).
Billy the kid and Pentecost mass (both classical AB recorded)
For those classical tracks, decorrelation increases in lower levels and decreases for higher levels.
(https://i62.servimg.com/u/f62/11/21/77/49/billy_10.png)
Bird on a wire and Fast car (both pop-rock)
(https://i62.servimg.com/u/f62/11/21/77/49/bird_o10.png)
Calculate the coherence function. Look at this https://www.dsprelated.com/freebooks/mdft/Coherence_Function.html
What you are doing is a home brewed approach to frequency dependent correlations.
Calculate the coherence function. Look at this https://www.dsprelated.com/freebooks/mdft/Coherence_Function.html
Thanks for the idea, I have to think about it.
Calculate the coherence function. Look at this https://www.dsprelated.com/freebooks/mdft/Coherence_Function.html
Thanks for the idea, I have to think about it.
You are welcome. Let me know how it goes.