> Quoting Ron (2014-10-04 02:52:24) > > On Fri, Oct 03, 2014 at 10:06:02PM +0200, Jonas Smedegaard wrote: > > If I do, I will refer to this bugreport - awesome explanations! > > ...please consider pass that explanation upstream!
We are forever tweaking this there based on feedback we get, indeed :) > > One other thing to be aware of, depending on how you're doing your > > comparisons, is that opusdec will dither by default, which improves > > the audible quality of low level signals, but does raise the measured > > noise floor. You can turn that off if you need to. > > Thanks, but sounds more detailed than the level I will be comparing at. > > Would be great in any case if such details was available somewhere - a > perhaps related one is that I have wondered if perhaps there was a > quality difference in downmix stereo to mono in melt or postpone it to > opusenc. They're definitely available *somewhere*, though you could indeed spend an entire lifetime collecting the pieces of the puzzle from where they are scattered! The --no-dither option of opusdec is at least not hidden, but knowing that dithering exists and why it is done is a little more occult. It does occasionally catch some people out when doing certain types of comparisons. If you haven't already seen them, I'd highly recommend watching the presentations here: http://xiph.org/video/ Episode 2 will show you (among other things) how dither will let you record a signal with less than 1 bit of dynamic range. Downmixing without losing quality is Tricky. In principle it's simple, in practice the best way to "preserve quality" is hugely dependent on your input signal. Since all downmixing is essentially summing some proportion of the channels, there's a very real problem that if the channels you are summing contain a signal that is 180 degrees out of phase, it will be completely cancelled out in the summation and all you'll get in the downmix for it is silence. This is true for both analogue and digital signals and is what noise cancelling headphones and the like deliberately exploit. It happens in the air when you have more than one speaker or microphone too, but if you have more than one ear that becomes a location cue for your brain rather than complete cancellation. For some input signals, what you lose is almost unnoticeable, but for others, entire pieces of the audio can go quite obviously mia, and the only way to 'fix' that is to process the signal to introduce a phase shift - which can then in turn make other parts of the signal (which are hopefully less noticeable) go missing instead ... Which really means the only sure thing is to try it, listen to it, and if the result sucks, break out a signal processor and mixing desk to try to manipulate it in ways that make it suck less. For Opus, you might actually find the best option is to just record it in stereo. The stereo encoding will happily exploit that sort of correlation as redundancy it can compress, so for the "same quality" a stereo file won't be twice the size of a downmixed mono version of it, and the decoder can extract a stereo file in mono without reapplying the stereo separation it encoded (though that also still has the possibility of phase cancellation occurring in some cases). In the end, it really all boils down to why you want it downmixed to mono in the first place, and what particular "quality" is the most important to preserve. You're inevitably going to lose something in the conversion, so the trick is to control what it will be. This is why we can't replace good sound engineers with very small shell scripts :) -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
