You can see that the noise isn't nearly as bad (these formats do not murder the data *that* much), and it's also clear that MP3 does not preserve the boundaries because it does not support seamless playback (which causes the horizontal shift and color change due to RGB mismatch).
I don't think whatever method you're using to convert the image to PCM samples is correct/useful. Here are my results for MP3, Vorbis, Opus, at default settings, with the image interpreted as 8-bit mono samples.pic.twitter.com/pIoq2yuSBk
-
-
-
Meanwhile in the Opus output you can clearly see the effect of the DC removal filter, which means solid areas tend towards grey.
- Show replies
New conversation -
-
-
I can't quite get my head around the shifted MP3... are the first n-samples from the input lost, then the next m-samples zeroed? (shift + black pixels) Is the darker Opus output due to some kind of automatic gain control or high-pass filter?
-
oh, same concept, different words I think...
https://twitter.com/marcan42/status/1121535962612297728?s=19 …
End of conversation
New conversation -
-
-
To both of you. Audio compression algorithms are for compress audio, not images. This experiment is ridiculous since most audio algorithms are based in how we hear audio, and not how we see images.
-
That's exactly what makes the experiment interesting.
- Show replies
New conversation -
-
-
Why does Opus look like you printed it directly onto metal? Looks cool actually...
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.