Strictly Stereo
Trade: Strictly Stereo
It could be argued that level compression is also a form of information reduction.
The sound of 320 streaming...
- Thread starter cooky1257
- Start date
In this context it is not.
Strictly Stereo
Trade: Strictly Stereo
How so?
Jim Audiomisc
pfm Member
In principle, provided you know the exact 'recipy' used to level compress you should be able to find an 'expansion' process that undoes the compression and restores the renders the same output as if no compression had been applied.
This trick is, nominally, what lays behind the HDCD method. The HDCD encoding squashes peaks and plays games with the gain. But adds a 'trail' of metadata which 'should' undo this given a suitable decoder.
However more generally in practice, that may not be possible because of a maths trapdoor called 'many to one'[1]. Even given the details of how level compression was applied, sometimes there is no practical reversability method.
So although level compression isn't a synonym for a form of information loss. It can certainly lead to it.
And of course, if the level compression is something you can't undo, then for all practical purposes the orginal information *has* been hidden from you, which means much the same as 'lost' if there is nothing you can do about it.
[1] Hard clipping is an example of 'many to one' mapping. A range of many input levels are all converted into the *same* output level. Given only the output you'd have to guess what the input actually was if you only have the clipped version. Sometimes you can guess reliably, sometimes not. But a guess means you didn't have the actual info.
This trick is, nominally, what lays behind the HDCD method. The HDCD encoding squashes peaks and plays games with the gain. But adds a 'trail' of metadata which 'should' undo this given a suitable decoder.
However more generally in practice, that may not be possible because of a maths trapdoor called 'many to one'[1]. Even given the details of how level compression was applied, sometimes there is no practical reversability method.
So although level compression isn't a synonym for a form of information loss. It can certainly lead to it.
And of course, if the level compression is something you can't undo, then for all practical purposes the orginal information *has* been hidden from you, which means much the same as 'lost' if there is nothing you can do about it.
[1] Hard clipping is an example of 'many to one' mapping. A range of many input levels are all converted into the *same* output level. Given only the output you'd have to guess what the input actually was if you only have the clipped version. Sometimes you can guess reliably, sometimes not. But a guess means you didn't have the actual info.
davidsrsb
pfm Member
ALAC is Apples version of FLAC and achieves ~50% bitrate reduction from 1.5Mb CD. Nobody manages significantly better than this without discarding information
...yes the concept of 'F'ree is a bit alien to 'A'pple...
darrenyeats
pfm Member
Yeah, FLAC is about 50%.
Jim Audiomisc
pfm Member
From Information Theory, if the data compression systems (e.g. flac or alac) is "efficient" then the result is an amount of data to store or stream that essentally equals the amount of information contained. The problems are:
1) Getting an "efficient" process means matching the method to the type of information.
2) Noise gets treated as "information" because the system can't usually tell random background noise from real wanted information. Hence the bloating of 'high res' data-compressed files and stream.
1) Getting an "efficient" process means matching the method to the type of information.
2) Noise gets treated as "information" because the system can't usually tell random background noise from real wanted information. Hence the bloating of 'high res' data-compressed files and stream.
davidsrsb
pfm Member
As the bit depth goes up, those extra bits become more and more like white noise and don't compress.
MP3 and AAC are basically trying to separate the music from noise and discard the latter.
Jim Audiomisc
pfm Member
Yes. Figure 2 on
http://www.audiomisc.co.uk/MQA/cool/bitfreezing.html
gives an example of just how severe this can be for some 'high rez' material. Hell of a lot of noise - which methods like flac and alac can't really compress at all. Hence my own conclusion that almost all 'high rez' material would be better off noise shaped down to 16bits per sample. Essentially the same real audio data without the bloat, and thus much smaller files. See figure 4 on
http://www.audiomisc.co.uk/MQA/intoshape/NoiseShapingHighRez.html
The problem being that people now equate "24bit" with "high rez" because almost no-one understands the use and practical results of dither and noise shaping. :-/
Most burgers are stuffed with fat to look like more meat.
Jim Audiomisc
pfm Member
Given that SACD/DSD is dithered *one bit per sample*, maybe they shouldn't. 8->
TBH My own view is that high rate 8 (or even 4) bits per sample should be just fine for high rez if done correctly. To me, it would make far more sense than SACD/DSD, but presumably wasn't patentable or sellable as a closed system. It would also have avoided the formal flaws of SACD/DSD.