GNOME Bugzilla – Bug 766218
MP3 decoding yields differerent results in 1.8 than 1.6
Last modified: 2016-05-10 11:49:01 UTC
I am not sure if it is a regression or not, but when I compare results of the the following command: gst-launch-1.0 filesrc location=Data.Record.File_a5ac9095-2a47-41e4-b746-a4d46da7f7fd ! decodebin ! audioconvert ! audioresample ! audio/x-raw,format=S16LE,rate=11050,channels=1 ! fdsink | wc -c for the same file between 1.6.2 and 1.8.1 I get 2 different results: 1.6.2: 4055201 1.8.1: 4053488 I would expect that MP3->raw decoding would have been quite deterministic.
Test file: https://pub.radiokit.org/gstbug766218/Data.Record.File_a5ac9095-2a47-41e4-b746-a4d46da7f7fd
Logs: https://pub.radiokit.org/gstbug766218/1.6.2.log.xz https://pub.radiokit.org/gstbug766218/1.8.1.log.xz
1) is the same decoder being used in both cases? 2) how does the output after the decoder compare? (to eliminating the effect of the convert/resample)
Also is the file size the same in both cases? "wc -l" is not a good metric for comparing in any case, some checksum would be more useful.
I've checked MD5, they are equal. Ok, on 1.8.1 there is fluendo MP3 decoder, I hadn't noticed one. That's the cause, sorry for bothering.