[gelo]

the codecs in most lossy compression formats usually nuke the phase information so here in this example as ASK is being used there isnt any lost of information. If he tried to encode phase changes in the sound then you wouldnt recover it since the decoder regenerates the audio with 0 deg phase.

[mistercow]

What do you mean by "nuke the phase information"?

[vardump]

Probably just that mp3 compression loses phase information? As it indeed does.

[mistercow]

Can you elaborate? What specifically about mp3 compression causes the phase information to be lost? Clearly, if you simply take IDCT(DCT(s)) for some signal s, you don't lose phase information, because the DCT is invertible. So are you saying that quantization causes the phase information to be lost? If so, why do binaural recordings still work fine when compressed as mp3s, given that they depend heavily on phase differences between the sounds reaching each ear?

[vardump]

> If so, why do binaural recordings still work fine when compressed as mp3s, given that they depend heavily on phase differences between the sounds reaching each ear?

That's a very good point. If phase information is lost, low frequency stereo sound localization should indeed suffer.

Also discrete cosine transformation indeed keeps phase.

I guess mp3 shouldn't be losing phase information after all.