compression codec comparisons

[Tom]

Interesting discussion on a private industry forum I frequent, a member made some comparisons between the different forms of compression used today, basically because some production companies want to use compressed source material, before its gone through the production chain (which will mean the signal gets converted through several formats, introducing lots of lovely artefacts to grate your ears).

Basically what he did was to have an uncompressed wav file, then convert it to mp3/wmv/whatever, then place the two files on top of each other and subtract, leaving only the compression artefacts. He then produced noise figures for each one (0dB is the reference level)

Atrac3+ at 255 kbps -37 dB
MP3 at 197 kbps (100% setting) -32 dB
Windows Media file at 160 kbps -25 dB
Atrac3 at 132 kbps (Minidisc-LP) -23 dB
MP3 at 92 kbps (25% setting) -19 dB

Its interesting stuff, because audio codecs try and disguise the part of the signal they compress, this comparison skirts around that and just gives pure figures. 92kb/s mp3 is the worst, atrac3+ is the best. WMV isn't doing too badly in the middle.

 

[Xavier]

I'm not sure what value the above "comparison" gives, he's changing too many variables at once...

If he were to test a series of bitrates under each codec, from say 64kbps to 320kbps then we would be able to work out some kind of pecking order, or alternatively to look at known codec/bitrate combinations which all conform to the same data rate(e.g. 1Mbyte/minute).

edit:// in fact by definition, it's not really a comparison at all, it's like comparing tyres when you've got set A on a Saxo and Set B on a ferrari... there's no defined metric

 

[phlash]

Hmmn,

Since MP3 works by removing those parts of the signal it rekons you won't notice (psychoacoustic processing), then using a bit compression scheme, you are bound to get significant waveform differences to uncompressed signal, what matters is if you can hear the difference, which is of course subjective (but can be tested using double blind methods)

Phil.

 

[Tom]

Well, not exactly. The point is that devices that use lossy compression are starting to appear within the production chain. Now at first generation, these codecs all sound quite good, but when you then push them through the production process, all kinds of very nasty things start to happen, because effectively you're eventually using several different codecs to compress the signal, which can have all kinds of horrid audible side effects.

From a quality point of view, its best to keep lossy codecs out of the equation until the final point of transmission. Try it, run an mp3 through successive codecs at the same bitrates, and listen to the results.

Xav I take your point about a poor comparison, I think the guy will do an exhaustive comparison one day, should be quite interesting to see the results.

 

[phlash]

Ah right - I somehow missed that bit about production chains Tom (too late in the day!) - fair point about multiple encodings causing increasing levels of distortion / audible artefacts - it's almost like going back to analogue infact