Judging quality by spectrograph

[Mephesto]

Is a studio quality spectrograph reliable enough to judge the quality of a codec/bitrate with than going thru the hassle of doing an ABX test? ABX is only good enough to judge up to transparency anyhow while a spectrograph can visualize quality beyond transparency.

Does anyone have a good example of a song looking really bad on a spectrograph yet being better quality than the one appearing better? The idiots at hydrogenaudio got their panties in a twist because I used a spectrograph to claim that AAC was way better quality at 192 kb/s than MP3 (as if an ABX test is somehow relevant for this) and confidently rebutted my claim with a sarcastic mention of BladeEnc MP3 being way better quality than AAC because it has higher frequency response (amazingly I'm somehow the troll here.)

This caught my interest so I compared AAC and BladeMP3 at the same bitrate. Blade actually did preserve more upper frequencies but the audio was rife with smearing and this was clearly visible on the spectrograph unlike with AAC that perfectly preserved the 0-16kHz shelf while content above that was selectively preserved and above a certain point completely cut.

When I pointed this out, I was banned.

So I ask of you guys if you have a more feasible example.

[Cornucopia]

A spectrograph only tells you that there is energy at certain frequency bands, not that the energy in question bears any GOOD relationship to the remainder of the signal. (e.g. odd or even harmonics in proper proportions)

So, NO: Although I don't have a quick examples to give you, it could certainly be possible for a spectrograph to be visually "richer" but be "grungier" than a sparser but cleaner (more transparent?) sounding alternative.

Intermodulation distortion is a good example: it "fills the intervening space" with stuff that just trashes the sound.

Also, as noted, a classic case of "smearing" is when there is non-linear phase delay (varying by frequency), usually due to poor filter construction. A spectrograph is BLIND to this kind of thing, because it only graphs pitch (y-axis), time (x-axis) and amplitude (intensity). Therefore, a pair of encodings which used the same source material, but where one had GOOD (phase-linear, e.g. FIR) filtering and the other had POOR (e.g. IIR) filtering would look IDENTICAL on the spectrograph (though with a minute amount of shift in the time, based on frequency), but the difference in sound would likely be quite noticeable.

I'm not really sure whether I vindicated you or them or neither with this answer. Truth is, testing that stuff is more complex than just a spectrograph can show.

Scott

[Mephesto]

Thanks for the articulate post.

I should amend a few things I should've included in the first post. First off, the source WAV should be always available so the spectrographs of the processed files can be compared to a spectrograph of the original and we know what it's supposed to look like, otherwise I could only rely on things like which one has a smaller shelf cut or which appears to be more smeared when the more harmonic one could be actually artifacted with flanging without me knowing it.

The smearing you mention IS visible because the harmonics look smoother and less rigid like they are on the source. Energy in frequencies surrounding the band is added with fluctuating intensity (noise), quieter harmonics neighbored around louder harmonics are gone or badly smeared etc.

Take a look at these (enable java):
WAV and AAC
WAV and BladeMP3
AAC and BladeMP3

WAV and AAC zoom
WAV and BladeMP3 zoom
AAC and BladeMP3 zoom
Notice how AAC perfectly preserves everything except the quiet noise while Blade smears everything together.

Intermodulation distortion is a good example: it "fills the intervening space" with stuff that just trashes the sound.

Filling quiet patches with audible noise is rather visible, I've seen Opus do this.

I'm not really sure whether I vindicated you or them or neither with this answer. Truth is, testing that stuff is more complex than just a spectrograph can show.

Well, would there ever be a case where two different codecs at a same bitrate or same codecs at a bitrate with a difference no smaller than 16 kb/s wouldn't reveal graphic evidence of more or less degradation on a spectrograph?

If someone gave me randomly-named outputs of different codecs and I was able to deduce their quality correctly with a spectrograph, would this be an unprecedented milestone that would make Hydrogenaudio shit themselves?