Audio Processing on the PS3 and Xbox 360

[-tkf-]

That's an odd place to put the blame. Normalization is at fault? You'd think PCM tracks would also be normalized. If not, then the difference he'd be hearing would be akin to the loss/masking you might get by turning the volume knob a little too much one way. Short of certain processing software/hardware having odd definitions of volume adjustment and normalization, it just seems like he's reaching for some totally mechanical cause in order to take his own subjectivity out of the picture.

Yes the first thing one should do is to doubt this reviewer, because.. maybe there is some actual truth to what he heard?

Why do every "nerd" find it so hard to believe that there can be a difference between uncompressed and lossless sound on a HiDef disc. YES i know that lossless is "like" zip, but we have no idea (at lest not yet) how the innerworkings of the decoder is, and much more important, how the encoder works. Add to this the speculation that putting the sound through a filter before the compression would yield better compression as opposed to just let it encode from the original PCM.

 

[kyleb]

YES i know that lossless is "like" zip, but we have no idea (at lest not yet) how the innerworkings of the decoder is, and much more important, how the encoder works.

Regardless of the encoding/decoding techniques used, by definition, lossless compression decompresses to the exact same data that was put into it.

Add to this the speculation that putting the sound through a filter before the compression would yield better compression as opposed to just let it encode from the original PCM.

Of course that will change the sound. It would be easy to demonstrate digitally as well. So no need for speculation here, have you seen any evidence of this?

 

[corysama]

On the processing sound, what about wave traversal through the scene taking into consideration materials? Ordinarily broad hacks are used to simulate environment. I imagine acoustic environment modelling will always be a back-seat choice and so never more demanding than the devs care to allocate processing resources, but even then, in your opinion can a sophisticated audio processing system be applied now at a level that we could happily consider as not needing to be improved on, and without eating too much processing resources?

No problem http://www.insomniacgames.com/tech/articles/1007/files/thall_occlusion.pdf

Most of the effect of audio occlusion is simple low-pass frequency filtering. If you want to get fancy then you do a little path tracing to figure out your DSP settings. The Xbox1 DSP could do filtering that was sufficient for basic environment modeling if you just put in the effort to tell it what to do. I wouldn't want to design a real-life opera house using the Xbox1 DSP chip, but for real-time simulation it is actually pretty hard to beat a sophisticated software filter running on an SPU.

 

[-tkf-]

by definition, lossless compression decompresses to the exact same data that was put into it.

This is completely new to me, really. /irony

 

[pipo]

Now you know why it's called lossless instead of lossy compression...

 

[ShootMyMonkey]

Yes the first thing one should do is to doubt this reviewer, because.. maybe there is some actual truth to what he heard?

The fact that he quips about the cost and quality of his equipment, and the fact that he's jumping into a comparison between raw and lossless compressed specifically looking for a difference means... yes, I'll doubt the reviewer first. The "truth" to what he heard is insofar as he believes he heard a difference -- but that proves nothing. You'd have a harder time getting the same result out of people who never bought expensive equipment and got the chance to listen on high-end equipment with no financial burden on their part. Similarly, if you have people who've spent thousands on an expensive sound system, and tell them to compare files on their own hardware and *look* for a difference, you'll get a "night and day" difference every time even if you play the same track and tell them that one play was compressed and one wasn't.

YES i know that lossless is "like" zip, but we have no idea (at lest not yet) how the innerworkings of the decoder is, and much more important, how the encoder works. Add to this the speculation that putting the sound through a filter before the compression would yield better compression as opposed to just let it encode from the original PCM.

Which in turn says that the reviewer is quick to place blame on one thing which he assumes to be the dividing line, as opposed to thinking his own decoder might be a source of the problem or that his own willingness to assume ahead of time that a difference is there because it has to be is problematic. Moreover, he finds a pretty weak foothold. If it was something like the preprocessing you mentioned to improve compression rates, that would be a more valid cause... though I'd like to see evidence that sound designers actually do this... (or at least do this and still manage to keep their jobs)

Besides which, all encoding differences can easily be removed from the equation, which leaves only the decoder and the listener, both of which are subject to flaws. And since the listener is human, it only follows that the listener is inherently the most flawed object in the chain.

 

[-tkf-]

The fact that he quips about the cost and quality of his equipment, and the fact that he's jumping into a comparison between raw and lossless compressed specifically looking for a difference means... yes, I'll doubt the reviewer first.

I read his "quips" about his above average gear more like a "yeah i got nice gear but it´s not over the top". You think the whale is half empty?

His review is backed up by other reviews as well, btw

 

[kyleb]

His review is backed up by other reviews as well, btw

Sure, but his claim isn't backed by any empirical evidence, which again would be easy to obtain if there were any to be found.

 

[Shifty Geezer]

Whatever the reviews say, scientifically and rationally there's no way lossless audio compression can inherently sound different than uncompressed audio. It's exactly the same data, and if sent through the same audio pipeline, will come out sounding exactly the same. If the audio data entering the audio pipeline as source material isn't the same, then the compression scheme isn't truly lossless, or wasn't the same starting material to begin. eg. If a system used 24bit uncompressed audio and 16 bit lossless compressed, or 48kHz uncompressed audio and 44.1kHz lossless compressed, even though the codec is lossless, the quality won't be the same because the source material isn't the same.

It's no different to viewing a simple 16 colour graphic in uncompressed BMP or lossless PNG. the data is exactly the same. If on a computer the PNG looks different to the BMP, it's because either it's not going through the same rendering pipeline, perhaps being viewed on a different monitor or through a different input, or it's because the data within the PNG didn't come from exactly the same source image as the BMP. The PNG compression system doesn't alter the data. It doesn't shift the brightness, hue, RGB value, or anything. Lossless audio compression, if truly lossless and not bugged!, will take the same 1s and 0s of a digital audio stream and record such that they are read back as exactly the same data. Any variation to the audio doesn't come from the codec.

It's for this reason that some are seriously sceptical of audiophile claims and attribute them to 'the human element' because the facts are that there can be no difference, and anyone claiming to be hearing a difference because of the codec isn't. The only possible explanations are :

1) Their test isn't valid, routing the different audio sources through different audio pipelines, or using different source material for the different test samples.
2) There is no difference and any reported differences come from a fault in perception.
3) That science doesn't truly understand the nature of digital data, and it's possible for the form a data is delivered in to affect it's real-world properties. eg. If you fold a letter into half three times and put it in an envelope and post it to someone, what they receive at the end isn't 'as nice' as if you sent the letter unfolded in a bigger envelope, even though the content is exactly the same. Thus if you 'fold' an audio stream to preserve all data but make it smaller, perhaps the nature of unfolding it even at the audio source level, somehow on a quantum, spectral, natural energy level, affects the nature of the audio so it's not as nice as if it was never folded? This theory covers both perspectives, that there can't be a difference and yet people report a difference. But I don't attribute much plausibility to this theory!

 

[fearsomepirate]

Three words that cause audiophiles to run in panic:

Double blind study.

 

[-tkf-]

Three words that cause audiophiles to run in panic:

Double blind study.

Yeah, if you read the reviews of some of the titles that includes lossless and and PCM tracks you will find some that did blindtesting..

Of course the results aren´t statistically valid and good enough to be accepted as "proof", especially not here since the conclusion is that PCM was preffered over the lossless compression.

 

[-tkf-]

Whatever the reviews say, scientifically and rationally there's no way lossless audio compression can inherently sound different than uncompressed audio.

I don´t think anyone here doesn´t understand the concept of lossless compression.

It's exactly the same data, and if sent through the same audio pipeline, will come out sounding exactly the same. If the audio data entering the audio pipeline as source material isn't the same, then the compression scheme isn't truly lossless, or wasn't the same starting material to begin. eg. If a system used 24bit uncompressed audio and 16 bit lossless compressed, or 48kHz uncompressed audio and 44.1kHz lossless compressed, even though the codec is lossless, the quality won't be the same because the source material isn't the same.

It's no different to viewing a simple 16 colour graphic in uncompressed BMP or lossless PNG. the data is exactly the same. If on a computer the PNG looks different to the BMP, it's because either it's not going through the same rendering pipeline, perhaps being viewed on a different monitor or through a different input, or it's because the data within the PNG didn't come from exactly the same source image as the BMP. The PNG compression system doesn't alter the data. It doesn't shift the brightness, hue, RGB value, or anything. Lossless audio compression, if truly lossless and not bugged!, will take the same 1s and 0s of a digital audio stream and record such that they are read back as exactly the same data. Any variation to the audio doesn't come from the codec.

It's for this reason that some are seriously sceptical of audiophile claims and attribute them to 'the human element' because the facts are that there can be no difference, and anyone claiming to be hearing a difference because of the codec isn't. The only possible explanations are :

1) Their test isn't valid, routing the different audio sources through different audio pipelines, or using different source material for the different test samples.
2) There is no difference and any reported differences come from a fault in perception.
3) That science doesn't truly understand the nature of digital data, and it's possible for the form a data is delivered in to affect it's real-world properties. eg. If you fold a letter into half three times and put it in an envelope and post it to someone, what they receive at the end isn't 'as nice' as if you sent the letter unfolded in a bigger envelope, even though the content is exactly the same. Thus if you 'fold' an audio stream to preserve all data but make it smaller, perhaps the nature of unfolding it even at the audio source level, somehow on a quantum, spectral, natural energy level, affects the nature of the audio so it's not as nice as if it was never folded? This theory covers both perspectives, that there can't be a difference and yet people report a difference. But I don't attribute much plausibility to this theory!

So far there hasn´t been any evidence or fact presented on WHY it sounds different. I think it´s plausible that there were compromises on the lossless tracks in regard to the original track being "pre baked" to over greater compression ratio to save discs pace and badwidth.