well, I took a look at the files. from my knowledge of shorten there are two things it does that flac doesn't do: 1. it estimates the mean of the signal for each block, subtracts it out and stores it separately. but this is pretty useless for the predictors that shorten uses as they are pretty insensitive to the mean (try different values of -m from 0 to whatever and note practically no difference in file size). 2. it tests for exploitable bitshift within a block; e.g. if all the samples in a block have 0 for the least significant two bits, it stores the bitshift (2) and shifts the signal down 2 bits. normally this is also useless for CD audio since even the LSB is not stationary and there is no exploitable bit shift. BUT, you have stumbled on some recordings where the LSB is 0 for much of the file. as a matter of fact, in the worse track (track 6) almost the entire signal in both channels is like that. taking a 15 bit-per-sample signal and just shifting it left one bit is worse than just adding 1 bit of noise, and as distortion goes may actually be more audible, so I'm not sure who engineered these tracks. I've never come across a sample like this which is why I thought it wasn't useful to add that functionality to FLAC... maybe if this is a common practice I should put it in. Josh --- Mark Powell <M.S.Powell@salford.ac.uk> wrote:>> yeah, could you put up the FLAC version of the >> worst track that is less than 20 megs compressed? >> (I'll have to grab it with a 56k modem). by worst >> I mean the one where shorten beats flac by themost.>> also: >> >> 1. what version of shorten are you using? > > v3.1 > > # shorten -h | head -1 > shorten: version 3.1: (c) 1992-1999 Tony Robinsonand SoftSound Ltd> >> 2. what command-line options for flac and shorten >> did you use on this track? > > Nothing fancy. > > $ shorten file.wav > > $ flac -V file.wav file.flac__________________________________________________ Do You Yahoo!? Get personalized email addresses from Yahoo! Mail - only $35 a year! http://personal.mail.yahoo.com/
On Thu, 15 Feb 2001, Josh Coalson wrote:> well, I took a look at the files. from my > knowledge of shorten there are two things it does > that flac doesn't do: > > BUT, you have stumbled on some recordings where the > LSB is 0 for much of the file. as a matter of fact, > in > the worse track (track 6) almost the entire signal in > both channels is like that. taking a 15 > bit-per-sample > signal and just shifting it left one bit is worse > than just adding 1 bit of noise, and as distortion > goes may actually be more audible, so I'm not sure who > engineered these tracks.See what you mean. I've just looked at those tracks in a hex editor. Wow, like you say they seem to have been engineered by Satan himself.> I've never come across a sample like this which is > why I thought it wasn't useful to add that > functionality to FLAC... maybe if this is a common > practice I should put it in.Hmm. Does seem like a waste of time. Although it does seem strange that shorten can out perform flac in these limited cases. When flac does get out of beta and starts to be adopted by the etree folks (and all of us) then you're going to get more and more people asking about this? I know my statistics teacher would shake his head at this conlcusion , but I've compressed around 50 albums with flac and found 1 where the LSB is 0. Even if it's 1 in 200, that's going to be lots of cases where shorten out performs it? But, hey I program too and this seems like a pain in the arse to code just for these limited cases :) Glad you got to the bottom of it though. Knew you would. Later. Mark Powell - UNIX System Administrator - The University of Salford Academic Information Services, Clifford Whitworth Building, Salford University, Manchester, M5 4WT, UK. Tel: +44 161 295 5936 Fax: +44 161 295 5888 www.pgp.com for PGP key
> > I've never come across a sample like this which is > > why I thought it wasn't useful to add that > > functionality to FLAC... maybe if this is a common > > practice I should put it in. > ... > I've > compressed around 50 albums with flac and found 1 > where the LSB is 0. Even > if it's 1 in 200, that's going to be lots of cases > where shorten out > performs it? > But, hey I program too and this seems like a pain > in the arse to code > just for these limited casesactually, it's pretty easy to code and not even a real speed penalty, but I would have to find a place in the stream header for the bitshift #, which is pretty tight right now. if I can I will probably add it. Josh __________________________________________________ Do You Yahoo!? Yahoo! Auctions - Buy the things you want at great prices! http://auctions.yahoo.com/