[PD] Pure Data Buffir~ object
Fred Jan Kraan
fjkraan at xs4all.nl
Sun Sep 6 17:15:59 CEST 2015
Hi Alexandre,
> Hmm, so you say the object will compute the maximum table size no matter
> if we're using just, say, 128 samples? Even if the arraysize of the
> impulse response is just 128? As if it would always consume more CPU
> even with the same smaller impulses?
No, only the number of samples specified by the length argument or inlet
is used. But as this is convolution, is is done for every sample in the
frame. So convolving a frame with 64 samples with a 4096 sized buffer
means 262144 calculations. In this case it is more efficient to
translate the frame to the frequency domain, process there and translate
it back.
>
> I wonder how things work internally with [FIR~].
As far as I can see, it does something similar. It doesn't use the FFT
route.
>
> It is true that convolution is quite expensive. In the frequency domain
> it is better, but it has it's own complications. The best solution is
> actually "partitioned convolution". In Pd we have [partconv~] for that,
> which has a couple of issues, but it is great. So you can use it for
> spending the same as a maximum of 256 buffer size for impulses that are
> 4096 or even greater.
>
> Unfortunately, Max does not have a partitioned convolution object, I
> think that there had to be one... but whatever...
>
> Anyway, I don't believe 4096 is such a costy even if doing this
> increases cmputation time. And we still have other options like
> [partconv~] and [FIR~] in Pd. But in practice I think we'll have
> [buffir~] as a better/more versatile object/option than [FIR~] in the
> end. So we have more to gain than lose with this.
Well, when it becomes possible, you can check it. That is the
educational aspect :-).
>
>> Hopefully coming week, because after that, I'll be busy
>> with work-related issues.
>
> Thanks! So, I can see how maintaining this library is not an easy task.
> I care about it a lot, as I've been recently using it to teach Max, and
> now that extended reached a dead end I believe it's one of the main
> libraries to be taken seriously and expand Pd. So I'm trying to get a
> friend of mine onboard with this, he's a teacher at a university and
> might be able to have students helping in a project. His name is Flavio
> and he's copied here. We were talking about writing a project to get
> some support for this project.
All help is welcome. Proper described and demonstrated issues with
objects are half the work of fixing them. There are unit-test patches in
the Pd-extended distribution which can be used for this.
>
> I keep finding issues with objects as I try them, so I still think there
> might be a lot more to fix. I'm about to write another email on issues
> with [matrix~]. And I also think we could clone some more objects.
> Hopefully this happens and we'll have more support to take care of this
> library, huh?
I have no intention to stop supporting cyclone. But it can happen there
are periods when time for it is limited.
>
> cheers
Greetings,
Fred Jan
>
> 2015-09-05 18:36 GMT-03:00 Fred Jan Kraan <fjkraan at xs4all.nl
> <mailto:fjkraan at xs4all.nl>>:
>
> Hi Alexandre,
>
> On 2015-08-30 08:41 AM, Alexandre Torres Porres wrote:
> > Howdy, hope you had a nice vacation buddy ;)
>
> Yes, thank you. It took some time to buffir~, but here we are :-).
> >
> >> There are no other serious issues with buffir~?
> >
> > haven't really played with it extensively, did try i t a bit and seemed
> > alright.
> >
> >> A first glance it doesn't look too complicated, but I think it would be
> >> good to retain the current buffer size as default, and add the larger
> >> size as an extra argument or message.
> >
> > I wonder why, since you already specify the table size in the object as
> > arguments. Doesn't make much sense to me adding up another argument if
> > I'm using a larger IR table. I guess the only difference would to change
> > the help file and say "max size is 4096" instead of "256".
>
> The second argument is the number of samples to be used in the
> convolution, not the size of the buffer, which is external. The first
> argument is an offset in the used buffer.
>
> >
> >> Maybe creating a separate object, like buffir4k~, somewhere
> >> outside cyclone would be the easier option.
> >
> > Why would it be easier? It wouldn't conflict. Once you had a newer
> > version working, that's it, no hard and no harm done.
> >
> > Maybe that concern seems like making a big effort to not expand the
> > buffir~ object, which I fail to see why. If you have a working version
> > that allows bigger tables, there's no compatibility issues or anything,
> > you're just able to get more out of it. So what would the concern be?
> >
> > But I also have to say that there already exists an external that allows
> > up to 4096 table sizes, that's [FIR~], from iemlib. So we wouldn't
> > really have to do that at all... And, yeah, I knew about [FIR~] before,
> > my request wasn't towards the necessity of a better and unavailable
> > external object that'd do that, it was just a concern to make [buffir~]
> > in cyclone as useful as I think it should be (let me stress and state
> > that I believe 256 is really too tiny and restricting).
>
> Buffir~ uses convolution in the time-domain, meaning using much larger
> buffer sizes isn't interesting performance wise, the local expert told
> me. Better performance can be reached using frequency domain filtering.
> But from the educational point of view, it is better to let the user
> decide which method is best, and change the BUFFIR_MAXSIZE value.
>
> Hopefully coming week, because after that, I'll be busy with
> work-related issues.
>
> Greetings,
>
> Fred Jan
> >
> > With the situation now that the development of extended ceased, and with
> > the great work you've been doing taking care of cyclone (and also making
> > it easily available into the deken plugin), I think all the objects in
> > cyclone need attention, even if there are alternatives to it. The way I
> > see [buffir~] now is that it is not useful, and that I'd rather use
> > [FIR~] instead. But I wish I could rely on cyclone for it without
> > needing to install a whole other library looking for alternatives of
> > features that an object from cyclone should have.
> >
> > I just think that if it isn't hard to expand it, and if it'd be working
> > after that, then no more worries :)
> >
> > what do you say?
> >
> > who else would be with me ;) ?
> >
> > cheers
> >
> > 2015-08-29 10:54 GMT-03:00 Fred Jan Kraan <fjkraan at xs4all.nl <mailto:fjkraan at xs4all.nl>
> > <mailto:fjkraan at xs4all.nl <mailto:fjkraan at xs4all.nl>>>:
> >
> > Hi Alexandre,
> >
> > > hey, how is going with this object?
> >
> > Not much, I just returned from vacation ;-).
> > >
> > > I just wanted to say it'd be really good if it was updated to max size
> > > of 4096, which is the current configuration since max 6
> > >
> > > 256 is just too little and made more sense back in the day of max 5
> > >
> > > hopefully it's an easy update, what do you say?
> >
> > A first glance it doesn't look too complicated, but I think it would be
> > good to retain the current buffer size as default, and add the larger
> > size as an extra argument or message.
> > Maybe creating a separate object, like buffir4k~, somewhere outside
> > cyclone would be the easier option.
> >
> > There are no other serious issues with buffir~?
> > >
> > > cheers
> >
> > Greetings,
> >
> > Fred Jan
> > >
> > > 2015-08-01 15:18 GMT-03:00 Fred Jan Kraan <fjkraan at xs4all.nl <mailto:fjkraan at xs4all.nl>
> <mailto:fjkraan at xs4all.nl <mailto:fjkraan at xs4all.nl>>
> > > <mailto:fjkraan at xs4all.nl <mailto:fjkraan at xs4all.nl>
> <mailto:fjkraan at xs4all.nl <mailto:fjkraan at xs4all.nl>>>>:
> > >
> > > Hi Julian,
> > >
> > > Thank you for your report. I am not the expert, just one of
> > the current
> > > maintainers if the cyclone set of objects. I think the
> object
> > works, but
> > > maybe not as the Max object. If this is the case I will
> try to
> > fix it.
> > >
> > > > Hello,
> > > >
> > > > I am sorry to bother you but I have an important question
> > about the
> > > > buffir~ object.
> > > > I hope you can help me with this. I am not too experiencd
> > with PD
> > > but I
> > > > try to make
> > > > an Binaural-Renderer. I want to use the buffir~ object to
> > convolve
> > > an IR
> > > > to a signal.
> > > > This works right now. But I want to change the IR used for
> > convolving
> > > > with the signal
> > > > when the head orientation changes. I had a look at the
> MAX/MSP 5
> > > > documentation
> > > > as I understand this object was taken from MAX where
> the buffir~
> > > object
> > > > can be told
> > > > to change the buffer~ object by using the set message. I
> > tried this in
> > > > PD but nothing
> > > > happens. Obviously the IR is not loaded as there is no
> > signal passing
> > > > through the
> > > > buffir~ object when using $1 as the name of the table
> and send a
> > > message
> > > > [set tablename]
> > > > to input 1 (the signal input where the clear message
> can be
> > > applied). Is
> > > > this function
> > > > not implemented in the PD buffir~ object? Or is there
> > something I
> > > do wrong?
> > >
> > > Note '$1' is not literary part of the name, just a
> placeholder
> > of the
> > > first element on the message (=list) you send to the
> object. If
> > > something is wrong here, the console should contain an error
> > message.
> > > Also see the help on $-variables.
> > > >
> > > > I would be verry happy if could help me with this
> question!
> > If this is
> > > > not implemented
> > > > and can't be worked around do you have any ideas how
> to solve my
> > > problem?
> > > > Thank you very much in advance!
> > >
> > > The attached patch is an adaptation of the help-patch,
> including a
> > > second buffer to switch to. Yo need to open it and draw
> > non-zero values
> > > to get a signal through. It looks like you also have to
> > provide the
> > > offset and length arguments, as leaving them out results
> in zero -
> > > meaning no useful convolution takes place.
> > >
> > > When I have time, I'll check the behaviour of the Max5
> object.
> > It might
> > > work better or more user friendly. But first on the planning
> > are some
> > > weeks vacation :-).
> > > >
> > > > Regards Julian Jochheim
> > > >
> > >
> > > Greetings,
> > >
> > > Fred Jan
> > >
> > > P.S. I added the pd-list in the reply, the main forum for
> > issues like
> > > this one (http://lists.puredata.info/listinfo/pd-list).
> > >
> > > _______________________________________________
> > > Pd-list at lists.iem.at <mailto:Pd-list at lists.iem.at>
> <mailto:Pd-list at lists.iem.at <mailto:Pd-list at lists.iem.at>>
> > <mailto:Pd-list at lists.iem.at <mailto:Pd-list at lists.iem.at>
> <mailto:Pd-list at lists.iem.at <mailto:Pd-list at lists.iem.at>>> mailing
> list
> > > UNSUBSCRIBE and account-management ->
> > > http://lists.puredata.info/listinfo/pd-list
> > >
> > >
> >
> >
>
>
More information about the Pd-list
mailing list