64 kbps listening test 2005
Reply #162 – 2005-08-19 05:10:17
Beware of the bias of testing only killer samples. Comparing codecs on killer samples will only give you a hint on how they perform... on killer samples. Not on regular music. See here for a thread on the subject.[{POST_SNAPBACK}][/a] keep your ridiculous ideas on conducting listening tests to your own thread. [a href="index.php?act=findpost&pid=319701"][{POST_SNAPBACK}][/a] Pieroxy's ideas have nothing ridiculous. He was very clear: a listening test involving VBR and CBR encoders should include low-bitrate encodings, mid-bitrate encodings and high-bitrate encodings. I'm [a href="http://www.hydrogenaudio.org/forums/index.php?showtopic=36184&view=findpost&p=319058]quoting[/url] Pieroxy:So in an ideal test, you would have 160kbps, 130kbps and 90kbps samples. THAT would be representative, not merely 128kbps. Now I'm quoting someone who's not used to post "ridiculous ideas" about listening tests, ff123 :Ideally, you'd like the bitrate distribution to look somewhat like a bell curve with its mean at 128 kbit/s. => Exactly the same "ridiculous" idea about the "ideal " test. And ff123 followed:For the 48 kbit/s test, if there are VBR codecs, I think we should strive to have about an equal number of bitrates above and below the average bitrate (which should work out to be 48 kbit/s on average across the sample set). Would you also say that ff123 has ridiculous ideas about listening test organisation? I'd really like to see that... I posted the bitrate distribution curve of VBR encoders involved in my latest 80 kbps listening test:http://foobar2000.net/divers/tests/2005.07...UTION_WMA80.png http://foobar2000.net/divers/tests/2005.07...ION_aoTuV80.png For both encoders, ~50% of the total samples were encoded with a bitrate lower to the targeted bitrate, and the ~50 remaining percent were encoded with a bitrate superior to the targeted bitrate. The distribution is close to the perfect curve for WMA and classical music (150 samples), with both mean and median at exactly 80 kbps . The curve is perfectible for aoTuV (median is 75 kbps), but distribution is also near-symetrical from both side of the median value . It was maybe luck - I don't know - or maybe a consequence of the huge number of samples involved in my test (150+35) which maybe help to get a good distribution.