On Mon, 29 Jul 2013 14:15:28 +0000 Fons Adriaensen <fons@xxxxxxxxxxxxxx> wrote: > On Mon, Jul 29, 2013 at 03:27:33PM +0200, Philipp Überbacher wrote: > > > while the paper > > (http://www.dolby.com/uploadedFiles/Assets/US/Doc/Professional/ > > AES128-Loudness-Normalization-Portable-Media-Players.pdf) > > is interesting, it's not what I was looking for. What needs testing > > is whether the loudness adjustment of RG2 compared to RG1 is > > perceived as 'closer to equal loudness'. When some sort of actual > > listening test shows that RG2 performs significantly better, good, > > otherwise I don't see the point. > > ITU-1770 on which both R-128 and RG2 are based has been tested > extensively over a long time and shown to work very well. AFAIK there > was no comparable amount of testing done on RG1. Really? I can't find any proper references, at least not in the latest version of the ITU-1770 paper (http://www.itu.int/rec/R-REC-BS.1770-3-201208-I/en). They only roughly describe some tests, first as mostly female speech (p. 13-14), later they claim it was a broad base of material (p. 18). I can't see any obvious reference to details of those tests, the second referenced paper (Evaluation of Objective Loudness Meters) seems to describe it, but it seems this paper is not publicly accessible. However, my point here is that it has to show its performance and especially superiority over the established RG1 in practice, everything else is of little significance. > > I do find their recommended conversion equation questionable, they > > came up with it based on their sample set plus thumb measure. > > There's no telling what the results will be, but I don't expect > > them to be any good. > > Why not ? The tests on the sample set (which is not small) show quite > a good agreement between the two systems. > > Ciao, According to the R-128 paper the standard deviation of their samples from the line they plotted there is 1dB, no idea what it is from the slope (-1) they guessed there. Still, given to songs it could well be 2dB or more off, but my main point is that this is still with _their_ data set. I guess that with another data set the differences could be larger and significant to the point where the error is large enough for the conversion to be a bad thing™. I wonder whether a better, albeit slightly more complicated way of conversion can be found, this one was apparently chosen primarily for its simplicity. Don't get me wrong, I like ReplayGain, and I'd like to see further adoption, I'm just not sure the R-128 based one will actually be better and cause further adoption. I fear it could cause confusion and further problems instead. Sometimes I also like to argue and see the bad parts in everything. On the upside, the temperature just dropped below 30° for a change. Have a nice day, Philipp _______________________________________________ Linux-audio-user mailing list Linux-audio-user@xxxxxxxxxxxxxxxxxxxx http://lists.linuxaudio.org/listinfo/linux-audio-user
