Op 03-12-14 om 16:48 schreef Olivier Tristan:> Looks like I've missed the talk about this regression introduced in 1.3.1.
Le 03/12/2014 17:30, Martijn van Beurden a ?crit :> Op 03-12-14 om 16:48 schreef Olivier Tristan: >> Looks like I've missed the talk about this regression introduced in 1.3.1. > >From > http://lists.xiph.org/pipermail/flac-dev/2014-October/005146.htmlok, in my use case, I should use "--best -e" instead of only "--best" then ? Thanks, -- Olivier Tristan Research & Development www.uvi.net
On Dec 3, 2014, at 8:30 AM, Martijn van Beurden <mvanb1 at gmail.com> wrote:> Op 03-12-14 om 16:48 schreef Olivier Tristan: >> This patch changes the settings associated with compression >> levels 6, 7 and 8. With this patch, -e is no longer used, but >> instead apodization functions are added. This should improve >> compression with at least 95% of all material. Independent tests >> show that this is probably the case. > > As it turns out, the material for which the compression didn't > improve is mostly 'quiet' material.I would like to point out that 'quiet' material is common when making original recordings. I do a great deal of live recording, and the levels are always quiet in order to guard against clipping. These live recordings are then archived via FLAC to make sure nothing is lost. There are also field recording units that record directly to FLAC, and they will also see relatively 'quiet' recordings compared to fully mastered commercial CD music. Even though those recorders might not be updated to the latest FLAC sources, I would hope that things are not getting worse for a whole category of FLAC users. Basically, I'm worried that the FLAC sources are being fine-tuned to handle only the subset of audio represented by commercial CDs, without regard for 24-bit or raw tracks that have not been fully mastered. It's interesting to note that the original FLAC algorithms performed quite well across the board, but recent 'improvements' are sacrificing performance in one area to enhance performance in another area. Are these the sorts of things that can be handled by expanding the compression options, rather than thwarting existing performance? Brian Willoughby
Le 03/12/2014 18:14, Brian Willoughby a ?crit :> On Dec 3, 2014, at 8:30 AM, Martijn van Beurden <mvanb1 at gmail.com> wrote: >> Op 03-12-14 om 16:48 schreef Olivier Tristan: >>> This patch changes the settings associated with compression >>> levels 6, 7 and 8. With this patch, -e is no longer used, but >>> instead apodization functions are added. This should improve >>> compression with at least 95% of all material. Independent tests >>> show that this is probably the case. >> As it turns out, the material for which the compression didn't >> improve is mostly 'quiet' material. > I would like to point out that 'quiet' material is common when making original recordings. > > I do a great deal of live recording, and the levels are always quiet in order to guard against clipping. These live recordings are then archived via FLAC to make sure nothing is lost. > > There are also field recording units that record directly to FLAC, and they will also see relatively 'quiet' recordings compared to fully mastered commercial CD music. Even though those recorders might not be updated to the latest FLAC sources, I would hope that things are not getting worse for a whole category of FLAC users. > > Basically, I'm worried that the FLAC sources are being fine-tuned to handle only the subset of audio represented by commercial CDs, without regard for 24-bit or raw tracks that have not been fully mastered. It's interesting to note that the original FLAC algorithms performed quite well across the board, but recent 'improvements' are sacrificing performance in one area to enhance performance in another area. Are these the sorts of things that can be handled by expanding the compression options, rather than thwarting existing performance? >It's the case actually Using -l 12 -b 4096 -m -e -r 6 will bring the previous --best behavior if understand correctly -- Olivier Tristan Research & Development www.uvi.net
Le 03/12/2014 17:42, Olivier Tristan a ?crit :> Le 03/12/2014 17:30, Martijn van Beurden a ?crit : >> Op 03-12-14 om 16:48 schreef Olivier Tristan: >>> Looks like I've missed the talk about this regression introduced in 1.3.1. >>> From >> http://lists.xiph.org/pipermail/flac-dev/2014-October/005146.html > ok, in my use case, I should use "--best -e" instead of only "--best" > then ? >nevermind found the info somewhere -l 12 -b 4096 -m -e -r 6 Thanks, -- Olivier Tristan Research & Development www.uvi.net
Op 03-12-14 om 17:42 schreef Olivier Tristan:> ok, in my use case, I should use "--best -e" instead of only > "--best" then ?If you don't mind the slower encoding, yes.
Op 03-12-14 om 18:14 schreef Brian Willoughby:> I would like to point out that 'quiet' material is common when making original recordings.This is probably a misunderstanding: I said that the material that suffers is usually quiet, not that all quiet material suffers.> I do a great deal of live recording, and the levels are always quiet in order to guard against clipping. These live recordings are then archived via FLAC to make sure nothing is lost.So do I, and I included a few unaltered samples with plenty of headroom in my standard test corpus for these improvements.> but recent 'improvements' are sacrificing performance in one area to enhance performance in another area.That is certainly not the case. The compression level retuning doesn't work beneficial for all material, but it is in this case simply coincidence that it is some quiet material that seems to suffer. The patch attached to this thread you're replying to however, for some reason benefits specifically this kind of sounds.