what does the exhaustive model search do?
Does it try all of the functions listed under "-A" to find the
'best',
i.e. bartlett, bartlett_hann, blackman, blackman_har-
ris_4term_92db, connes, flattop, gauss(STDDEV), hamming, hann,
kaiser_bessel, nuttall, rectangle, triangle, tukey(P), welch.
Or is that something else?
I've been using the -e option for a while in my flac encoding,
but for something that sounds like it should be slow I wonder
if it is testing all the models or if it might not be
working correctly?
Thanks...
On Jun 10, 2012, at 21:32, Linda A. Walsh wrote:> what does the exhaustive model search do? > > Does it try all of the functions listed under "-A" to find the 'best', > i.e. bartlett, bartlett_hann, blackman, blackman_har- > ris_4term_92db, connes, flattop, gauss(STDDEV), hamming, hann, > kaiser_bessel, nuttall, rectangle, triangle, tukey(P), welch. > > Or is that something else? > > I've been using the -e option for a while in my flac encoding, > but for something that sounds like it should be slow I wonder > if it is testing all the models or if it might not be > working correctly? > > Thanks...A better question might be: "What do the -A options do?" All of those windowing functions are lossy, and are used for frequency domain transforms. I'm not sure how they would be used in a lossless encoder. Then again, I have not yet studied the entire mathematics behind FLAC. As for your question, I've never used the exhaustive model search option, but I had originally assumed that it meant a search among the various compression level options. For example, -l -b -r -M and -m all are preset when using -0 through -9 and --fast and --best as if they were macros, but you can also manually set those options in different combinations. I initially thought that the exhaustive search went through the -l -b -r -M/-m options to find an adaptive "best" compression rather than a preset one. However, now that you've made me think about this in more detail, I tend to assume that the exhaustive model search has more to do with the LPC (linear predictive coding). The key to lossless compression is to find a model that predicts each sample from the previous samples, and the better the model the smaller the file. An exhaustive model search must go through all possible LPC models rather than the quicker default list. Anyway, my apologies for responding without an actual answer, but a conversation might be slightly more interesting than a quick answer (I hope). Brian Willoughby Sound Consulting
Brian Willoughby wrote:> On Jun 10, 2012, at 21:32, Linda A. Walsh wrote: > >> what does the exhaustive model search do? >> >> Does it try all of the functions listed under "-A" to find the 'best', >> i.e. bartlett, bartlett_hann, blackman, blackman_har- >> ris_4term_92db, connes, flattop, gauss(STDDEV), hamming, hann, >> kaiser_bessel, nuttall, rectangle, triangle, tukey(P), welch. >> >> Or is that something else? >> >> I've been using the -e option for a while in my flac encoding, >> but for something that sounds like it should be slow I wonder >> if it is testing all the models or if it might not be >> working correctly? >> >> Thanks... >> > > > A better question might be: "What do the -A options do?" > > All of those windowing functions are lossy, and are used for > frequency domain transforms. I'm not sure how they would be used in > a lossless encoder. Then again, I have not yet studied the entire > mathematics behind FLAC. > > > As for your question, I've never used the exhaustive model search > option, but I had originally assumed that it meant a search among the > various compression level options. For example, -l -b -r -M and -m > all are preset when using -0 through -9 and --fast and --best as if > they were macros, but you can also manually set those options in > different combinations. I initially thought that the exhaustive > search went through the -l -b -r -M/-m options to find an adaptive > "best" compression rather than a preset one. > > However, now that you've made me think about this in more detail, I > tend to assume that the exhaustive model search has more to do with > the LPC (linear predictive coding). The key to lossless compression > is to find a model that predicts each sample from the previous > samples, and the better the model the smaller the file. An > exhaustive model search must go through all possible LPC models > rather than the quicker default list. > > > Anyway, my apologies for responding without an actual answer, but a > conversation might be slightly more interesting than a quick answer > (I hope). > > Brian Willoughby > Sound Consulting > > _______________________________________________ > Flac mailing list > Flac at xiph.org > http://lists.xiph.org/mailman/listinfo/flac >-------------- next part -------------- An HTML attachment was scrubbed... URL: http://lists.xiph.org/pipermail/flac/attachments/20120612/58fb4ab0/attachment.htm
Linda A. Walsh
2012-Jun-13 22:16 UTC
[Flac] repost of correct response... flac -- exhaustive model search vs. -A <*>?
I just noticed that a bad copy of this went to the list (bad in that it didn't include my comments to Brian...*oops*) -- rectified.. Brian Willoughby wrote:> On Jun 10, 2012, at 21:32, Linda A. Walsh wrote: > >> what does the exhaustive model search do? >> >> Does it try all of the functions listed under "-A" to find the 'best', >> i.e. bartlett, bartlett_hann, blackman, blackman_har- >> ris_4term_92db, connes, flattop, gauss(STDDEV), hamming, hann, >> kaiser_bessel, nuttall, rectangle, triangle, tukey(P), welch. >> > A better question might be: "What do the -A options do?" > > All of those windowing functions are lossy, and are used for > frequency domain transforms. I'm not sure how they would be used in > a lossless encoder. Then again, I have not yet studied the entire > mathematics behind FLAC. >----------------- ?!?! really? Lossy?... I can't see how that would fit into a FLAC format?... (logically). [I mean, how can one use those and still call yourself a 'Lossless' format?]> As for your question, I've never used the exhaustive model search > option, but I had originally assumed that it meant a search among the > various compression level options. For example, -l -b -r -M and -m > all are preset when using -0 through -9 and --fast and --best as if > they were macros, but you can also manually set those options in > different combinations. I initially thought that the exhaustive > search went through the -l -b -r -M/-m options to find an adaptive > "best" compression rather than a preset one. >---- Well, anyone who hasn't tried the exhaustive model search, should -- it's darn fast. I can't imagine any of the higher numbered options being much faster. [as collections (album/cd or multiples of each), VERY often are perfect for parallel encoding (making most albums encodeable in well under a minute, if not <10-15 seconds on a 2-physical-cpu, system (8-12 cores total).> However, now that you've made me think about this in more detail, I > tend to assume that the exhaustive model search has more to do with > the LPC (linear predictive coding). The key to lossless compression > is to find a model that predicts each sample from the previous > samples, and the better the model the smaller the file. An > exhaustive model search must go through all possible LPC models > rather than the quicker default list. > > Anyway, my apologies for responding without an actual answer, but a > conversation might be slightly more interesting than a quick answer > > (I hope). >--- In absence of hard facts, theories always tend to come first... It's more scientific sounding than the ones with hordes of daemons exhaustively trying every answer... ;-)