Hello again. I discovered a disappointing, to me, bug in j2kaudio/libjasper today while I was working on 32-bit audio support.
My previous holy-grail-looking lossless audio compression test results were incorrect. I was passing rate=1 to the JasPer JPEG 2000 encoder, which was not performing lossless compression, but slightly lossy compression. I actually do not know what it was doing because the behavior I was getting is somewhat undocumented. I could not differentiate the quality by ear; everything sounded perfect to me. I only discovered this when I had to do sample by sample checks for accuracy and I found that they were very close, but not exactly the same. The results matched exactly when I left rate= out of the encoder options entirely, but the resulting file size increased. I will post a question to the jasper email list to find out what rate=1 should be doing and why it does not match omitting the rate. The new lossless results from my test song are: la - 65.5% ofr - 66.3% ma - 66.9% lpac - 69% flac - 70.2% shorten - 75.2% j2kaudio - 76.6% (dead last) bzip2 - 96.7% gzip - 97.1% Thank you again for all of your feedback and input. I apologize for getting anyone's hopes up and for not performing a sample-by-sample check at the integer level sooner. For now, I think I have proved and improved the concept, but I conclude that even though it outperforms file compressors, other lossless and lossy audio compressors are superior for audio data. Updated tables and information: http://geocities.com/hildstrom/projects/j2kaudio Thanks. -Greg Gregory Alan Hildstrom Secure Systems Engineer - Trusted Computer Solutions trustedcs.com [EMAIL PROTECTED] Software Engineer - Hildstrom Engineering geocities.com/hildstrom [EMAIL PROTECTED] Mobile:(210)413-6082 Fax:(210)599-0469 Office:(210)340-3151x117 _______________________________________________ Linux-audio-dev mailing list Linux-audio-dev@lists.linuxaudio.org http://lists.linuxaudio.org/mailman/listinfo.cgi/linux-audio-dev
