Skip to main content

Topic: Comparing the 'difference' of encodes (Read 1878 times) previous topic - next topic

0 Members and 1 Guest are viewing this topic.
  • Mac
  • [*][*][*][*][*]
Comparing the 'difference' of encodes
Sometimes I check out the differences in audio codecs on a sample, and because my ears can't ABX anything well, I need to rely on visuals to spot the differences.

It's been hammered into me that spectographs cannot judge quality of a psyco-accoustically encoded file.  But, I'm wondering if there is a valid use.

Basically, I take the original wave and the encode, and subtract one from the other (making sure they're perfectly in time).

I suppose this gives me the exact information the codec threw away, which I can then listen to. 

Is it reasonable to deduce things from this discarded info?  I suppose really you can only tell the quality when it's taken from the original, as it depends how well the lost info is masked by the remaining stuff.  I would at least think you can infer the amount lost, and any points where it has produced a noticable deviation from the original sound...  both things I can't usually tell from just listening to the original
< w o g o n e . c o m / l o l >

  • JohnV
  • [*][*][*][*][*]
  • Developer
Comparing the 'difference' of encodes
Reply #1
Quote
I suppose this gives me the exact information the codec threw away, which I can then listen to.  

Is it reasonable to deduce things from this discarded info?  I suppose really you can only tell the quality when it's taken from the original, as it depends how well the lost info is masked by the remaining stuff.  I would at least think you can infer the amount lost, and any points where it has produced a noticable deviation from the original sound...   both things I can't usually tell from just listening to the original

No, that's not a good way to compare codecs. Every codec does the masking a bit differently. What counts, is the audible quantization noise, not the total quantization noise. Of course, when you substract the encoded from the original, you get the total quantization noise.

Of course there's sometimes, like if there are large peak deviations, correlation between the overall quantization noise and audible quality. Then again, even if the total quantization noise looks overall low amplitude and all, it can still fail if the masking is not right. So even if there's less total quantization noise, there can be more audible quantization noise, and that's what counts in audible quality.

This is not a safe way to judge quality between codecs. I'd guess this might be better if you test one codec (one psymodel) with different settings. But even then I wouldn't count on it in every case.
  • Last Edit: 07 December, 2002, 08:06:42 AM by JohnV
Juha Laaksonheimo