Re: Seedeclip4 Mediaplayer entry deleted, edits undone and user blocked. Why?
Reply #31 – 2017-01-10 11:07:19
This discussion is trying to accomplish too many things at once. Substantive debate about declipping Back-and-forth about the loudness wars and statistics of distributed music Debate about the proper ways to deal with overly broad statements on the wiki and other disputes there Questions about policies for allowing software authors and other experts to contribute to the wiki Questions about Greynol's behavior patterns as a moderator and a wiki editor Questions about cutestudio's edits and whether he should have wiki edit privileges At least some of these discussions need to be separated somehow and addressed more generally and dispassionately. CuteStudio, if you think greynol has overreacted and acted irrationally or based on emotion/hot temper, the most important thing for you to do is to avoid overreacting, act rationally, and keep temper and emotion in check. Try to be more concise and dispassionate in your complaints. Here's my take on the first few of those topics; I may comment on the others later. 1. In a couple of other threads as well, Greynol and others have been unjustly dismissive of declipping. For highly variable signals such as speech, where clipped regions are likely to be short, isolated, and very severe, good audio restoration algorithms frequently improve the SNR by more than 12dB. It is not at all difficult to hear the improvement either (esp. the reduction in boomy full-spectrum distortion during clipped vowels). For less variable signals (most music), if the severity of clipping is high enough to make an obvious audible difference, it's likely that so many samples are clipped that a restoration algorithm has insufficient information for a really good reconstruction. Improvements are likely to be much smaller, say 6dB or less, but may still be worth pursuing. This is not some kind of weird audiophile junk. This is a well-posed set of mathematical problems which have seen good theoretical and engineering work. Sadly the only open-source stuff I'm aware of are Audacity's clipfix, which is a very naive (cubic interpolation is not very appropriate for audio) simple hack Ben Schwartz (later of Xiph fame) did back as an undergraduate, and Monty's postfish, which does a reasonable job but is not as simple to work with (linux only, no distro includes binaries, source only available via svn, odd build dependencies, doesn't integrate into other kinds of toolchains). Audacity's "repair" effect, which does least squares autoregression, could probably be turned into a halfway decent declipper if combined with something that detected the clipped regions. Implementing the algorithms from any of the various recently published academic papers on the subject might be more competitive with the closed-source solutions. 2. If someone puts an overbroad statement about clipping in pop songs on the wiki, the right way to deal with that is a  and then remove the offending statement if evidence is not provided in a timely way. We don't need to sit here and argue about the loudness wars and how they should have been addressed twenty years ago. 3. I don't think mass deletion and reversion and blocking users at the first suspicion is sensible for dealing with a small closed-membership wiki. Thanks for your thoughts jensend, all relevant and all noted. Greynol hasn't come back to me yet to explain any of his actions, I can only assume he's on holiday. Censorship is always difficult to defend and to counter, it seems to be more and more the 'answer' to awkward subjects that no one likes discussing. Moderation is usually a way to keep a forum/wiki running smoothly, one has to question it's effectiveness when it does the exact opposite, stifling discussion and suppressing alternate viewpoints - even when they are backed up with evidence and proof. I'm also learning that the quality of the digital source is a sacred cow here. For me as a programmer it's simply input data so I have no bias - I'm used to all types of data quality, perhaps the audio hobbyist is particularly attached to a romantic 'perfection' of the signal. Clues for this are in the number of posts about 'bit perfect' quality. Boy if they looked at their 'bit perfect' waveform in Audacity they'd have a nasty surprise!! This isn't helped by people never seeing the waveform - but I have to admit it's disappointing that people on a hobby audio website are too afraid to even look for themselves. This is a typical modern track with around 50,000 clips. The histogram shows quite severe compression (rather than limiting which is a bulge like a dolphin surfacing). That histogram would be very sparse - mostly black - on a well recorded track, that bright white is all artificial compression. Note there is still clipping, so it was also given a good shave after the compression. Click on the thumbnail/blob to view. Perhaps it's all too familiar for me as I see it every single day. Literally - in the detail above. Note that this song is not 'special' for modern pop, it's average - I've seen (and mistakenly paid for) much worse. One can lead a horse to water but making it drink is another ball game.