Ogg Vorbis Development Progress
Reply #56 – 2005-03-04 17:52:28
I have also taken a look to this complementary block switching algorithm. To do this, I’ve called a full Panzer Division of samples. Some of these samples are maybe not relevant: instead of using only “micro-attacks” samples, I have also put some “micro-details” ones. What’s the exact difference? I can’t explain it well. By microattacks I mean many short, repeated and contiguous attacks; by microdetails I mean one of a few short and soft attacks or simply musical asperities. Microdetails are similar to microattacks, but with lesser intensity. They’re not necessary detected as something sharp, and therefore, are sometimes neglected by the encoder. Microdetails are therefore removed, or blurred and they disappear (partially or totally) after encoding. I’ve noticed in the past that vorbis tends to soften and sometimes remove these microdetails. All encoders tend to do this, but vorbis apparently more than lame, mpc or aac. Rival offers generally a crisp and sharp sound. Vorbis could be compared to RV9 or RV10 video format, which “overblurs” some fine details, whereas other formats could be compared to MPEG-4 encoders which try to maintain these details. I expect from a new or complementary block-switching algorithm to solve this issue, as well as thicker asperities called “microattacks”. I have limited the test to ~130 kbps encodings (-q4): it’s much easier and faster to test than higher bitrate encodings; and this bitrate corresponds to my current interest for lossy formats (average bitrate and highest quality possible). All samples are uploaded, and therefore could be tested by anybody interested in higher bitrate. I’ve compared 1.1 encoder to the modified one (QK 2005.02.28 now called QK28). It might be interested to put results in perspective by comparing them with another high quality format. Just to see if I’m not requesting from Vorbis results which are not achieved by competitors. I have decided for AAC. The nature of these samples (difficult one, with microattacks) requires a VBR method to maximize quality. That’s why Nero AAC was used instead of iTunes AAC, very good too, but CBR only. Results are on this post:http://www.hydrogenaudio.org/forums/index....showtopic=32080