JohnSwenson wrote: > First let me say that I don't have a fully defined set of rules such as > "if xyz is greater than abc it sounds bad" . Also it's difficult to find > out exact implementation details of particular filters without the > designers telling you what they did in the chip. (some do, most don't) > > What I have done is looked at the graphs in the spec sheets and listened > to many different chips and come to some conclusions, then came up with > a number of hypothesis and run a bunch of tests to see how these pan > out, and have come up with some theories based on that. Note these are > just observations, I have no idea WHY it is this way. > > Number one theory is the shape of the passband. If it looks > "complicated", different frequency bands looking very different, or > looking like two different simple functions superimposed on each other, > then the sonics are compromised. All such filters are combinations of > more than one type of filter. This seems to be the number one issue, > every test I have done shows that a single simple filter sounds better. > Exactly WHAT filter is not nearly as important as it being a simple > filter. > > The problem is that in order to get very high stop band attenuation in > reasonable amount of silicon takes such complex filters. And for some > reason the stop band attenuation has become THE metric companies are > trying to optimize in their spec sheets. The result is that all embedded > digital filters are complex types, and hence are sonically compromised. > My opinion is that this is why there is so much talk about "all the > chips sound alike", they are all being similarly compromised by these > complex filters. > > One interesting aspect of this is that many chips have a complex filter > for 44.1 and 48, a less complex filter for 88.2 and 96 and a basic > simple filter or no filter at all for 176.4 and 192. I think this is why > a lot of people go for upsampling. Unfortunately if you do it in > hardware, the upsampling is done with a complex filter, so you haven't > gained much, BUT by doing it in software you can use a high quality > simple filter. This difference in filter complexness is probably part of > why there is such interest in the higher sampling rates. > > Theory two, ultimate stop band attenuation is not all that critical. > With no filter you can definitely hear the aliasing, I like to > characterize this as the sound is "dirty". BUT you don't have the issues > of the complex digital filters. I think this is why NOS (No Over > Sampling) has a significant following. To get good sound you DO need to > get rid of the aliasing, but how much? I have found that 60db of stop > band attenuation is perfectly acceptible to clean up the sound, and you > can easily achieve that with a simple filter. I have yet to hear a > difference with any simple filter just by changing the stop band from > 60db to 120db. But huge differences when switching to a complex filter > to get that 120db. > > That's it for the theories. I have a bunch more observations but haven't > had enough testing to get into theories. I have done some listening to > things like zero phase, minimum phase, pre ringing, post ringing, long > time, short time etc. They do make differences, but I have not come to > any conclusions about what sounds best. The differences these things > make are small in comparison to the difference between a complex and a > simple filter. > > There has been a lot of talk about the "steepness" of the filter, my > experience is that in and of iteslf this is not very important, the big > issue is that again in order to get a very steep filter most > implementations use a complex filter. The "gentle slope" filters are > usually much simpler and THAT it the real difference. > > Again no hard data on this, it's just me listening to a whole bunch of > stuff and trying to come up with some correlations. Also no clue WHY > human hearing seems to be so sensitive to complex filters. > > John S. Thanks John I'm intrigued to try this out, but am not sure how to be sure that i am getting a simple rather than complex filter. When upsampling in sox what filter would you recommend - would I be safe specifying a Ilinear phase filter cutting in at say 18 kHz?
------------------------------------------------------------------------ adamdea's Profile: http://forums.slimdevices.com/member.php?userid=37603 View this thread: http://forums.slimdevices.com/showthread.php?t=97881 _______________________________________________ discuss mailing list discuss@lists.slimdevices.com http://lists.slimdevices.com/mailman/listinfo/discuss
