I just did listening test of lossless->opus100kbps->opus100kbps using early Alice in Chains track (demanding because of distorted guitars and dynamically mixed drums) and it sounds good enough. Some artificialness is hearable if you listen closely, but it's not disturbing.
W dniu 2014-02-08 04:33, Daniel Jonsson napisał(a): > Won't that require a lot of disk space though? And what should I do > if > all I got, for example, is an mp3 file? The short answer: It will require a lot of disk space but it's necessary if you want good sound quality. Storage is cheap nowadays (as long as you have your own server, though...) The best options is to not accept lossy compressed files. Demand FLAC. If they don't know what FLAC is, demand WAV. If you have CD, rip it yourself using cdparanoia or EAC, to FLAC. If you download music (who doesn't? ;) ) add "FLAC", "lossless", "APE" or "ALAC" to your search query. If you get music from Jamendo, use my script jamendo-dl: https://github.com/adiblol/audiotools/ If you _REALLY_ want to source from lossy, make sure artifacts aren't hearable _AT ALL_ in source files (though good headphones or monitors) in comparison listening test. If they are, re-encoding will make them even worse and more disturbing. And do you _REALLY_ have to use so low bitrate for your music storage? The long answer: Lossy audio codecs use perceptual coding which uses properties of our hearing such as masking to discard what we don't hear. They are designed to work with uncompressed audio. If you feed lossy codec with lossy audio (lossy-to-lossy transcoding), it will degrade quality much more than lossless-to-lossy transcoding. See http://wiki.hydrogenaudio.org/index.php?title=Transcoding for more information. I am frequently being shocked by songs on youtube sounding like shit so I tested whether Vorbis (used by YouTube in WebM mode) 130kbps is really that bad. I used lossless source and it sounded near transparent. Then I tested MP3 128kbps and artifacts weren't obvious and disturbing, only barely hearable. And recently I listened to some track on YT. It was only 96kbps but sounded better than some YT uploads tagged HD with 192kbps. Probably the band uploaded it in good quality, unlike unaware people uploading mp3 to youtube... So lossy codecs can sound very good even at low bitrates, if encoded properly!!! In theory it is possible to design lossy audio codec that doesn't rely on perceptual coding but it would be less effective. I know 2 modern codecs that use such approach: * WavPack (lossy mode) http://en.wikipedia.org/wiki/WavPack#Hybrid_mode * LossyWAV - http://wiki.hydrogenaudio.org/index.php?title=LossyWAV Music encoded this way could be transcoded to vorbis/opus/aac/whatever... without quality loss. If I understand correctly, LossyWAV uses dithering and reducing bit depth so it only increases noise floor (as Compact Cassette / 8track did) but doesn't introduce artificially sounding distortion. Unfortunately LossyWAV is written in Delphi and runs only on M$ Crap but works with Wine. It was the theory that audiophiles and sound engineers usually agree with and use in their work. What about practice? Some perceptual codecs bear transcoding better that others. MP2 and Musepack (which is based on MP2) are examples. If your source file has high bitrate (such as MP3 320 or Vorbis q>8), probably nothing bad will happen when transcoding it to another lossy format. The best advice is listening test. Do you know some golden-ear audiophile or good sound engineer (NOT semi-deaf club mixer operator!!!)? Ask them whether distortion is hearable. It is isn't, you are doing it right. That's all. Happy hacking! adiblol Self-appointed audio conversion expert (haha) ;) ------------------------------------------------------------------------------ Managing the Performance of Cloud-Based Applications Take advantage of what the Cloud has to offer - Avoid Common Pitfalls. Read the Whitepaper. http://pubads.g.doubleclick.net/gampad/clk?id=121051231&iu=/4140/ostg.clktrk _______________________________________________ Savonet-users mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/savonet-users
