Do radio People Understand Compression?

[chriscollins]

AAC+ is becoming a new standard, for sure. 64 kbps AAC+, when processed well sounds better than 128 mp3.

You will find many disagree with this statement. It will be better with regards to "swishies" for sure.

<- - snip -->

I base this on the fact I have never clued into the "synthetic quality" of the high end despite having spent _lots_ of time listening to AAC+ (during product development) on headphones and good speakers. On the other hand, I know of several people who's opinion I trust (Goran and Cornelius for example) who not only hear it, but find it very annoying.

You are blessed!

Maybe it won't happen to you, though I suspect it eventually will... Once your brain "clues in" on the artificially synthesized high-end that always has the same texture to it, you will hear it every time you listen to HE-AAC. The higher bitrate helps to mask this a little bit, but it is essentially unavoidable with HE-AAC simply because the way the codec works.

Perhaps it's easiest to hear on the music that you really know well and listened to it on CD, that has a very nice and soft cymbals or hi-hat. If you are used to the original sound of the high-end and listen to it coded with HE-AAC (AAC+) even at 64 kbps or 96 kbps, you will hear the high-end that's completely different than the original. It will have harsh, metallic and "gritty" sound to it.

And if you play another song that in the original (WAV/CD) has a different texture of the high end, after being encoded with HE-AAC (AAC+) you will hear exactly the same texture of the high-end as in the previous song with soft hi-hats. That's the problem of HE-AAC. By design it's not transparent and it will never be and the high-end always has the same "gritty" texture to it.

AAC, on the other hand, can sound transparent and faithful to the original if used with at least 128 kbps.

HE-AAC creates an artificial high-end, which is how it can achieve wide frequency response at very low bitrates. That's what it was designed for. It was not designed to give high audio quality. It was designed to give wider frequency response at very low bitrates.

The marketing people and some manufacturers that use HE-AAC in their products, have then created this claim of 48 kbps or 64 kbps HE-AAC v2 sounding as good as _________ (fill in with whatever you consider good quality) to make a strong selling point. They've claimed (and still do) better than analog FM quality. Which is how 48 kbps HD Radio channels came about.

But it's not true and long-term listening will prove it. Just ask anybody who has listened to any form of digital radio that uses SBR technology in the encoder - HD Radio in USA, DAB+ in Australia, HE-AAC encoded satellite radio, HE-AAC encoded web streams...

We need to use higher bitrates (like 128 kbps) and codecs such as AAC in digital transmissions, to claim any level of audio quality that will match analog FM or CD quality.


Regards,
Goran Tomas

Goran,

You are so much like me from a listening standpoint. I can hear artifacting even in high bitrate scenarios. After my 6 month 'trial' of Sirius in my new vehicle, I cancelled because I could only listen for 30 minutes at a time tops...

I can also see compression in Dish, H.264, etc... I guess some of us are just real sensitive to it.

I had someone encode two songs of their choosing at 320K MP3 and double blind test me ten times. I picked the mp3 out of every single one. I guess it is a curse...

 

[Tom Wells]

Another hearty agree with Goran.
A replicated high end response means an even greater stipping of info has occured, and the 'replicant data'
always has exactly the same eerie sheen, like Madame Tussad's wax dummies.

Has anyone else ever been to a wax museum like Ripley's?

That's what happens to audio in bad perceptual compression schemes.

 

[satech]

Another hearty agree with Goran.
A replicated high end response means an even greater stipping of info has occured, and the 'replicant data' always has exactly the same eerie sheen, like Madame Tussad's wax dummies.

Remember "MP3 Pro"? Back in the days of dialup Internet, when even a 128 kbps MP3 file was often considered too long of a wait to download, they came up with the idea of adding SBR to 64 kbps MP3, then claiming that it sounded equally as good as a regular 128 kbps MP3 file -- which, of course, was a lie. The MP3 Pro codec never caught on, but the SBR part of it was adopted into MP4/AAC.

I think the best way to describe SBR is to liken it to taking a color movie, copying it as black & white, and then using an artificial "Ted Turner" colorizing process to restore the colors. The result may look OK, but it is not, and never will be, identical to the true color original!

 

[gtomas]

I think the best way to describe SBR is to liken it to taking a color movie, copying it as black & white, and then using an artificial "Ted Turner" colorizing process to restore the colors. The result may look OK, but it is not, and never will be, identical to the true color original!

That's actually a very good analogy!


Regards,
Goran Tomas

 

[RolfTaylor]

Another hearty agree with Goran.
A replicated high end response means an even greater stipping of info has occured, and the 'replicant data' always has exactly the same eerie sheen, like Madame Tussad's wax dummies.

It is not so much that more info is removed. The core codec (e.g. AAC) is actually run a at half the normal sample rate. That is why an AAC decoder can decode HE-AAC/AAC+. Of course the high end is then handled completely separately.

This is also why there is so little improvement at higher bit rate. The portion handled by the core codec sounds better (less swishies and other usual codec artifacts) but the SBR portion still sounds the same.

Remember "MP3 Pro"? Back in the days of dialup Internet, when even a 128 kbps MP3 file was often considered too long of a wait to download, they came up with the idea of adding SBR to 64 kbps MP3, then claiming that it sounded equally as good as a regular 128 kbps MP3 file -- which, of course, was a lie. The MP3 Pro codec never caught on, but the SBR part of it was adopted into MP4/AAC.

Actually, AAC+ came first. The reason for MP3 pro was that it was backwards compatible with existing MP3 players yet sounded "better". As the whole discussion of Mp3 versus AAC earlier on this thread (or was it another recent thread?) this was an interesting approach as it did not force the need for a new decoder.

I think the best way to describe SBR is to liken it to taking a color movie, copying it as black & white, and then using an artificial "Ted Turner" colorizing process to restore the colors. The result may look OK, but it is not, and never will be, identical to the true color original!

Not sure if my comments above will show up properly....

Back when I was in graduate school in psychology I learned the following. Most studies of perception, sensation, etc, are geared towards discovering the norm. However, there is a whole field of "individual differences". There are a huge number of different factors that can be measured, and while many are highly correlated to each other, there are many that are independent of other traits.

As I mentioned before, I believe the type of listening studies done on codecs are geared towards the average performance and does not factor consider individual difference. So, while it is a rare ability to be able to pick out the 320 kbps MP3 versus the source (and for this reason I do not dispute Chriscollins' claims) I think it is much less of a rare ability to hear SBR. In other words, I believe there is a wider variance listening tests of SBR codecs even if the average performance is still "acceptable".

This is why I mentioned the fact that "many would disagree" with the claim that AAC+ is "better" than AAC.

Since I am no longer developing codecs, my opportunities to learn the SBR artifacts, are less. However I do listen to HD radio in the car and hope I will be able to continue to do so!

In an unrelated irony, I also discovered during product development that my voice is rather hard to code without artifacts!

 

[gtomas]

It is not so much that more info is removed. The core codec (e.g. AAC) is actually run a at half the normal sample rate. That is why an AAC decoder can decode HE-AAC/AAC+. Of course the high end is then handled completely separately.

This is also why there is so little improvement at higher bit rate. The portion handled by the core codec sounds better (less swishies and other usual codec artifacts) but the SBR portion still sounds the same.

Exactly! Rolf is right on the spot.

I didn't want to get too much technical (I was actually saving it for an article that would explain all this ;-)) but that's essentially it. With SBR, the core codec (whether it is AAC in HE-AAC, or MP3 in MP3 Pro) runs at half the sample rate. Meaning, it's encoding the spectrum only up to 8, 11 or 12 kHz. The SBR then re-creates the upper half of the spectrum.

And this is exactly why the bitrate increase with HE-AAC doesn't help much. And why, if you recognize the artifacts of the SBR technology, you can hear it with 48 kpbs and you can hear it with 96 kbps as well. And why HE-AAC can never sound transparent and provide high audio quality.


Regards,
Goran Tomas

 

[bigtom101]

Goran,

You are so much like me from a listening standpoint. I can hear artifacting even in high bitrate scenarios. After my 6 month 'trial' of Sirius in my new vehicle, I cancelled because I could only listen for 30 minutes at a time tops...

I can also see compression in Dish, H.264, etc... I guess some of us are just real sensitive to it.

I had someone encode two songs of their choosing at 320K MP3 and double blind test me ten times. I picked the mp3 out of every single one. I guess it is a curse...

I was pleased as punch to give Cablevision the heave ho and get FiOS because the pixellation drove me insane, especially on sports. The PQ on FiOS is second to none.

Hopefully both fixed and mobile broadband can offer increased speed and higher data caps as time goes on so that broadcasters (and listeners) aren't handicapped by bandwith limitations. Whenever I go thru a 4G service area in Central Jersey or NYC/Nassau County, my 128k stream loads as fast as if I'm sitting at home.

'till then...best we can do is try and mask it as best as possible with good processing.