Quote:
Originally Posted by jruggle
1) If FFmpeg is modified to give floating-point output will the results be closer to eac3to.
|
I don't know how to modify that patch to make it work with ffmpeg. If I apply it as is, ffmpeg outputs noise. So maybe someone else can check it?
Quote:
Originally Posted by jruggle
3) See what the Nero decoder does compared to liba52 and libavcodec.
|
Nero make it even more confusing. Here are
files. *-eac3to-nero.flac were decoded with eac3to; *-nero.flac with Nero 10 decoder and Graphedit.
Graphs:
So, eac3to-nero "looks" like ffmpeg for 5a.ac3, like liba52 for 224.ac3 and like eac3to-libav for 640.ac3.
I was surprised even more, when I compared SNR results:
Code:
5a-eac3to-libav: 27.194 dB (Gain for File B = 1.0009)
5a-eac3to-nero: 28.607 dB (Gain for File B = 0.99991)
5a-ffmpeg: 29.057 dB (Gain for File B = 0.99994)
5a-nero: 28.625 dB (Gain for File B = 1.5873)
5a-foo-liba52: 28.822 dB (Gain for File B = 0.99985)
224-eac3to-libav: 16.188 dB (Gain for File B = 0.99912)
224-eac3to-nero: 16.18 dB (Gain for File B = 0.99182)
224-ffmpeg: 13.399 dB (Gain for File B = 1.2648)
224-nero: 13.4 dB (Gain for File B = 2.0076)
224-foo-liba52: 16.166 dB (Gain for File B = 0.99174)
640-eac3to-libav: 27.809 dB (Gain for File B = 0.99982)
640-eac3to-nero: 27.813 dB (Gain for File B = 0.99982)
640-ffmpeg: 16.091 dB (Gain for File B = 1.2752)
640-nero: 16.091 dB (Gain for File B = 2.0243)
640-foo-liba52: 27.537 dB (Gain for File B = 0.99971)
Also, I've got 5a.ac3 decoded with Cyberlink PowerDVD 10.
5a-cyber.flac:
SNR = 28.844 dB (Gain for File B = 1.5872)
As you can see, Cyberlink hasn't that problem with attenuation of high frequencies.
Can anyone check these samples with Windows 7 decoder? Or maybe even with reference decoder?
Quote:
Originally Posted by jruggle
What program did you use to make the graphs? I want to use it in my comparisons as well.
|
Sound Forge. AFAIK baudline can provide
same functionality.