Post by Steve Forte Rio on Jan 2, 2014 19:37:54 GMT
Hm, interesting, why do we call it "reference" while it's standalone project, not an Apple or FhG product. If so there is no reasons of why to not use foobar2000 built-in (ffmpeg, AFAIK) decoder, as it surely supports decoding to 32-bit float for each codec.
The past public listening tests and my private listening tests slightly differs in the way the normalizations are done. Public - unweighted RMS, using the ABC/HR for Java Private - replaygain weighted RMS, using my rawgene4
There will approx. 40-50 samples. 85% - music (equal amount of samples per genre) 15% - non-music (speech, mixed as speech+background, ambient etc)
Steve Forte Rio has made a proposition here The samples will be picked randomly online rolz.org/group
Kamedo2, Steve Forte Rio and me have something on our mind. So the final desicion will be a fusion of our contributions.
My proposition to include an equal amount of samples from these sources:
1. The set of samples which was made by Chris. Link. It's well balanced set of samples. I'm in favor to include this whole set. 2. The samples from Hydrogenaudio members Link and MPEG sequences from here 3. A big list from previous test. Link
Predicted from the last 2011 AAC public listening test. We can split the job into two, to avoid fatigue on each donator. Chris's set have 16 samples, and we need to include the new samples people uploaded in HA, and the MPEG sequence.
My proposition: Whole set of samples from Chris ( 16 ) Whole set of samples randomly picked from the libraries of bitrate calibraters ( 12 ) 7 samples randomly picked from the MPEG sequences. ( 7 ) 7 samples randomly picked from the huge set of mainly hard samples, used in the last 2011 AAC test. ( 7 ) 8 samples randomly picked from the new set of samples from the HA thread. ( 8 ) Total ( 50 )
Of cource, the number is debatable. If many people agree, we can reduce them to all 8, and makeit 40 in total.