• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!

Information theory and sound quality

sweetmusic

Member
Joined
Apr 23, 2022
Messages
57
Likes
32
I was watching Amir's excellent (and long :) video about how Harman and other sound engineers determine objective and subjective sound quality. I'm pulling out my question to a separate thread because it's a different topic.

My question: what causes information *loss* from speakers, specifically?

Thinking about the frequency response curve and spinorama: if there are non-linearities, or dips and peaks, they can in theory be corrected, by equalization or by our brains by getting used to the deviations. No information is lost.

A bookshelf speaker that doesn't reproduce any sound below 45 Hz has lost all the low frequency information. It can't be recovered, though to a certain extent we might hear overtones and mentally fill in what the fundamental tones might be.

Similarly, distortion and noise actually cause information loss. It can't be recovered with fancy EQ, or by our brains.

What else besides limited frequency range, distortion and noise can cause information loss?
 
OP
S

sweetmusic

Member
Joined
Apr 23, 2022
Messages
57
Likes
32
Information theory: There's a digital music recording. The goal of perfect audio reproduction would be for the 0s and 1s to be transmitted losslessly to the listener so the sound can be reconstructed and experienced. Each stage of the music playback potentially introduces entropy. Distortion, noise, and limited frequency extension cause irretrievable information loss. Other information transformations are not lossy, like frequency deviations.

I studied information theory mostly in the context of compression algorithms and transformations. The psychology of perception gets interesting in audio, since some kinds of information loss are more critical than others. That's why AAC and MP3 drop high frequencies that many people can't hear and devote more bits to the presence region that people are more sensitive to.

Just trying to start a conversation and hear from people who know much, much more than me :)
 

Killingbeans

Major Contributor
Joined
Oct 23, 2018
Messages
4,089
Likes
7,547
Location
Bjerringbro, Denmark.
Thinking about the frequency response curve and spinorama: if there are non-linearities, or dips and peaks, they can in theory be corrected, by equalization or by our brains by getting used to the deviations. No information is lost.

Depends on whether the dips and peaks are consistent throughout the dispersion pattern. Otherwise you risk conserving information in the direct sound, while causing even grater error in the reflections.
 

DVDdoug

Major Contributor
Joined
May 27, 2021
Messages
2,920
Likes
3,834
Thinking about the frequency response curve and spinorama: if there are non-linearities, or dips and peaks, they can in theory be corrected, by equalization...
Dips can be a problem. A +6dB boost is 4 times the power and +10dB is 10 times the power. Bad-dips in the bass range caused by standing waves can require almost infinite power and infinitely-large woofers/subwoofers to overcome, so these are best-corrected with bass traps to kill the reflected waves.

A bookshelf speaker that doesn't reproduce any sound below 45 Hz has lost all the low frequency information.
Usually it will have some output but it can be too low (in volume) to hear or to be useful. Again with EQ you run-into power limitations or you can run-into mechanical limitations and/or drive the speaker into distortion or burn it out.

or by our brains by getting used to the deviations.
Your brain can play tricks but it can also "reset" when you listen to something with full frequency response. A long time ago I was working on a car stereo that had distortion... After awhile I started questioning if I was really still hearing distortion and I had to listen to something "clean" to "reset" my ears & brain. It was weird but interesting!

No information is lost.
The information is lost even if you don't notice it. ;)
 
OP
S

sweetmusic

Member
Joined
Apr 23, 2022
Messages
57
Likes
32
@DVDdoug good point about how your brain can reset is "speaker correction" after hearing a more accurate sound.

@Killingbeans is information lost in the off-axis, or is it just EQ'ed a little differently? Equalization in theory doesn't add or remove information, just levels things out, IIUC.

Hmm... In an information-theoretic sense, the frequency response doesn't matter as long as it's audible. All the information is there, even if it's shifted up or down a few decibels. Maybe that's the answer though... if the FR is low, there will be cases where it drops below the threshold of audibility, or the noise floor of the room, and then information is truly lost to us.

That's interesting... is that what happens at low listening levels as well, that sonic information is lost?

Do some speakers play lower volume sounds that others don't? Is there distortion or noise at extremely low sound levels that differs by speaker? In the measurements, I usually see distortion measured at 90 dB or other nearly ear-damaging levels. Do we understand what happens at more like 20-30 dB? Is there no distortion, but there's a complete dropoff to silence at some frequencies?
 

Killingbeans

Major Contributor
Joined
Oct 23, 2018
Messages
4,089
Likes
7,547
Location
Bjerringbro, Denmark.
@Killingbeans is information lost in the off-axis, or is it just EQ'ed a little differently? Equalization in theory doesn't add or remove information, just levels things out, IIUC.

The problem is that in many speakers the off-axis response can have dips and bumps that aren't on the on-axis response. If you level out on-axis you'd might cause new or greater exaggeration off-axis.
 

sigbergaudio

Major Contributor
Audio Company
Forum Donor
Joined
Aug 21, 2020
Messages
2,639
Likes
5,397
Location
Norway
Uneven frequency response may not technically be considered "information loss" but it's certainly information distortion. What is played back is inaccurate compared to what is stored on the recording. And as mentioned above it can't necessarily be "fixed" by EQ due to inconsistencies in on/off-axis response or inherent problems with the drivers. Inaccuracy due to other sources like cabinet resonances etc also can't necessarily be resolved.
 

sigbergaudio

Major Contributor
Audio Company
Forum Donor
Joined
Aug 21, 2020
Messages
2,639
Likes
5,397
Location
Norway
The problem is that in many speakers the off-axis response can have dips and bumps that aren't on the on-axis response. If you level out on-axis you'd might cause new or greater exaggeration off-axis.

Exactly. And unless the speakers are very cheap, it's reasonable to assume that the manufacturer has already considered this and made a reasonable trade-off between on- and off-axis response. Assuming that one can do this better than the designer of the speaker seems slightly arrogant (but may of course be possible in some cases). :)
 
Top Bottom