New Research on Audibility of Distortion in Headphones

[Doltonius]

0 dbSPL at 1000 Hz is the same as 50 dBSPL at 31 Hz! You go below 50 dBSPL @31 Hz and you can't even hear it. So by definition, bass is recorded at far higher amplitude than higher frequencies.

The equal loudness curve gets less dramatic as you get louder, no?

So by definition, bass is recorded at far higher amplitude than higher frequencies.

This is assuming that it is ideal for music to have comparable loudness at 30hz vs higher frequencies. Is this true? It could be that humans don't like too much bass compared to the mids and treble.

By 4 kHz, you are 35 dB down. And nearly 50 dB at 20 kHz. Again, this is a female, instrumental track.

When I play this track on Apple Music (yours is hi-res, so there might be production differences, but it can't be this big), the spectrum looks completely unlike the one you post. The following is what I see typically, and corresponds much better to what it sounds like (indeed, not a bass heavy track at all):

 

[amirm]

When I play this track on Apple Music (yours is hi-res, so there might be production differences, but it can't be this big), the spectrum looks completely unlike the one you post. The following is what I see typically, and corresponds much better to what it sounds like (indeed, not a bass heavy track at all):

You are taking a snapshot. I am showing you the energy across the full track.

 

[Doltonius]

You are taking a snapshot. I am showing you the energy across the full track.

I fear the way you do it distorts the situation, in that the higher frequencies are usually narrow spectral peaks that shift around a lot, and lower frequencies are wide spectral bumps cramped in the narrow band, and so if you average over time, you seem to not get a lot of energy at higher frequencies. But, at any moment in the track, dynamic peaks in mids are just as large as the bass peaks, which means in fact you aren't hearing the bass at much louder than the mids. To be able to hear bass at 100db and everything else at healthy levels, you need the vast majority of snapshots to look like your graph. But that is just not the case.

 

[azzy_mazzy]

The equal loudness curve gets less dramatic as you get louder, no?

This is assuming that it is ideal for music to have comparable loudness at 30hz vs higher frequencies. Is this true? It could be that humans don't like too much bass compared to the mids and treble.

When I play this track on Apple Music (yours is hi-res, so there might be production differences, but it can't be this big), the spectrum looks completely unlike the one you post. The following is what I see typically, and corresponds much better to what it sounds like (indeed, not a bass heavy track at all):
View attachment 533286View attachment 533282

what software are you using for this? using the track off youtube and viewing the live spectrum through foobar2000 the result im seeing match the graph posted by amir

 

[amirm]

To be able to hear bass at 100db and everything else at healthy levels, you need the vast majority of snapshots to look like your graph. But that is just not the case.

That makes no sense. I can have a track with drums alone to start. And then some with highs mixed in. The former does not in any shape have to look like my graph.

Anyway, you are wasting our time since it is clear you made an assertion with no research of your own. I am not here to do your homework. If you think high frequencies have equal to higher amplitude than bass, then the onus is on you to show a corpus of music following this.

 

[Doltonius]

what software are you using for this?

I am using ozone's parametric eq plugin, which has a live spectrum.

the result im seeing match the graph posted by amir

can you post it?

 

[Doltonius]

Anyway, you are wasting our time since it is clear you made an assertion with no research of your own.

Isn't the research just looking at spectrums of music, which I did?

I think I found the real problem, which is the spectrum plugin I looked at (iZotope Ozone EQ) has a built-in up-slope, which makes the spectrum of most music look flat. Switching to a spectrum set to no slope does show much bigger bass to mids contrasts, even in snapshots; the idea about not using spectral energy averaged over time should still hold. Your track now looks like this, still not as dramatic as in the average energy graph (and so not a good example of "bass 30db more than mids")

Here is a very bassy track (electronic anisong where bass absolutely dominates)

Now this might be really nearly 20db difference between bass and mids (vocals).

 

[amirm]

Your track now looks like this, still not as dramatic as in the average energy graph (and so not a good example of "bass 30db more than mids")

Peak at bass frequencies is around -27 dB. Peak at 3 to 5 kHz where our hearing is most sensitive is -53 dB. Difference is 26 dB. As I said, this is a vocal and not all a bass heavy track. So if this can manage this much difference, clearly there are tracks that have far larger differential.

Here is a very bassy track (electronic anisong where bass absolutely dominates)

Not correct. Techno type tracks have heavy upper bass, not deep bass where you can most of the time feel it, than hear it.

 

[Doltonius]

Not correct. Techno type tracks have heavy upper bass, not deep bass where you can most of the time feel it, than hear it.

This particular track has main bass peak at 40-60hz, as you can see in the image, I think this is subbass rather than upper bass, though indeed there is also quite some upper bass in addition to that

 

[solderdude]

I think I found the real problem, which is the spectrum plugin I looked at (iZotope Ozone EQ) has a built-in up-slope, which makes the spectrum of most music look flat.

Many analyzers for music have a pink-noise weighted display.
This looks more like how we perceive the sound.
The track fast car has a 'pink noise slope' to it.
Sometimes when I find a recording that does not sound tonally balanced I look at the spectrum and use EQ to get it closer to pink noise. Usually it sounds much better then.

Pink noise also is used in analyzers when used for speaker measurements. This ensures tweeters won't blow and sonically does not sound terrible and is similar to actual music in spectrum.
The display is contra pink noise weighted so that a flat line is seen on the display as if white noise were used. The opposite of pink noise is called blue noise.
So most analyzer displays in stereo systems are blue noise weighted so we can see an, on average, flat line of frequencies dancing up and down that looks like it sounds but does not reflect the actual voltage levels per frequency.

 

[olieb]

Many analyzers for music have a pink-noise weighted display.
This looks more like how we perceive the sound.

The question here is “What is flat?“
A FFT spectrum shows white noise as flat. This means constant energy per frequency/bandwidth. So the energy in 20-40Hz is equal to the energy in 2000-2020Hz. Only 1% of the energy in the 20-20 000Hz range is below 200Hz.
Is this the right kind of flat?
If one would measure energy per octave/relative bandwidth, then white noise would be shown as we hear it: hiss, no meaningful bass, all treble.
And pink noise, where the energy in 20-40Hz is equal to the energy in 2000-4000Hz, would be flat.
As the bandwidth of (most) sounds and of the filters (and hair cells) in our inner ear increase with frequency, it seems reasonable to me to use a spectrum per octave.
Otherwise a typical sound in bass will probably fall into only a few bins whereas a similar sound in high frequency (with same SPL and relative bandwidth) will fall into many bins, thereby showing a deceivingly low level,

 

[solderdude]

The question here is “What is flat?“
A FFT spectrum shows white noise as flat. This means constant energy per frequency/bandwidth. So the energy in 20-40Hz is equal to the energy in 2000-2020Hz. Only 1% of the energy in the 20-20 000Hz range is below 200Hz.
Is this the right kind of flat?

Yes.
Both for electronics as well as a speaker on axis in anecoich conditions should measure as well as a measurement microphone.

Acoustic 'flat' differs per SPL and then there is preference and taste on top of circumstances.

 

[olieb]

Both for electronics as well as a speaker on axis in anecoich conditions should measure as well as a measurement microphone.

That is about handling a given signal.
For an amp, transducer, etc, one wants that the signal is getting through without change (constant transfer factor for sound pressure, voltage, sample value) of course.
But that is true for white-->white (flat to flat for energy per bandwidth) as much as for pink-->pink (flat to flat for energy per relative bandwidth).
So for modifying (or not) a signal both measurements are equivalent (as long as you stay consistent).

The characteristics of a signal from a source is a different thing. It is a description of the signal not of the transfer.
The way to calculate a spectrum is a choice. You can use a way where “flat“ means that a (broadband) signal has hundred times less energy in bass octave 20-40Hz compared to the octave 2-4kHz (and a tenth of the sound pressure).
But if you do so, you should take that into account when judging amplitudes.

The key factor is the signal’s bandwidth characteristics. For isolated sinusoids (sweeps) both measurements give the same result. A constant amplitude sweep measures flat in both cases. But when the tone becomes a sound with a bandwidth the distribution into multiple bins starts to create a difference.

 

[solderdude]

Flat remains flat. Regardless of the tonality of music instuments.

 

[MRC01]

It seems to me that even though the high frequencies have much lower amplitude, when the musical signal also includes higher amplitude low frequencies (as it almost always does), the highs are small ripples surfing along on the much bigger bass wave. This means whatever membrane this wave hits (like a microphone diaphragm or an eardrum), that membrane vibrates similarly. The low frequencies cause large slow displacements, and the high frequencies are ripples riding on that.

Thus if you listen to that treble alone, the eardrum is barely moving, no risk of hearing damage. But if you listen to the entire waveform (including the lows), that treble at the same quiet level may be nearing the amplitude limits of the ear, which can cause hearing damage.

The situation is the reverse of IMD testing in a single-drive full-range speaker like a typical headphone. You can't detect the IMD with a tone sweep because it never plays 2 frequencies simultaneously. But you play a low and high frequency simultaneously, the distortion of the high tone increases dramatically, even though it is quiet, because the high amplitude of the low frequency tone pushes the driver to much larger excursions.