The Sound Of (lossy) Music

[Tenson]

I thought it might be interesting to hear only the added distortions introduced by different lossy compression methods.

This can be done by subtracting an original file and a lossy compressed version to leave only the artefacts of the compression.

I have compiled a file (HERE) as follows:

Original *BEEP* 128 MP3 *BEEP* 320 MP3 *BEEP* 320 Ogg *BEEP* Original

When I say '128 MP3' etc.. I actually mean just the artefacts. It may seem sensible that the file type leaving the least sound must be the best, and I do think the Ogg is clearly superior, however also consider that if the remaining sounds are nothing like music and horrible to listen to, they will be more obvious than pleasant artefacts at the same level. In this case though, the Ogg file leave so much less sound after subtraction with the original file it is laughable. No wonder high bitrate Spotify sounds good!

As a side note I used FhG MP3 encoder in high quality mode. However I think better can be had from mp3. I did a similar test with a 320 MP3 downloaded from Amazon and it nulled more like the Ogg file (although still not so good). I wonder if it was actually an Ogg file in am MP3 wrapper

 

[RobHolt]

This is one of a series of excellent videos mainly looking at loudness, but this one touches on lossy compression and includes some good examples.

 

[Werner]

When making presentations like this it is useful to emphasise that this is about perceptual coding and thus that the error signal heard in isolation is not very representative for the damage done: it is, according to the model used, supposed to be (near-)inaudible in the presence of the full signal.