Hello,

I've been playing around a little with the convolution opcodes. They  
seem very CPU demanding and at sr 96khz it is almost impossible to run  
long stereo IR files in real-time. I'm thinking of making a plug-in  
with csladspa for ardour. Has anyone got any experience with these  
opcodes? What is the most efficient and at the same time high quality  
opcode to use? Would it be better to simply multiply two pv-streams  
with the pvs opcodes?

Thanks in advance.

Best,

Peiman


Send bugs reports to this list.
To unsubscribe, send email sympa@lists.bath.ac.uk with body "unsubscribe csound"

Hi,

Have you tried pconvolve and different values of ksmps?

They should all give you the same quality, unless there's something
wrong with them.

I'm not sure using the pvs opcodes will work, as they are windowed. Or
will they?

Cheers,
Andrés

On Mon, Feb 8, 2010 at 4:13 PM, Peiman Khosravi
 wrote:
> Hello,
>
> I've been playing around a little with the convolution opcodes. They seem
> very CPU demanding and at sr 96khz it is almost impossible to run long
> stereo IR files in real-time. I'm thinking of making a plug-in with csladspa
> for ardour. Has anyone got any experience with these opcodes? What is the
> most efficient and at the same time high quality opcode to use? Would it be
> better to simply multiply two pv-streams with the pvs opcodes?
>
> Thanks in advance.
>
> Best,
>
> Peiman
>
>
> Send bugs reports to this list.
> To unsubscribe, send email sympa@lists.bath.ac.uk with body "unsubscribe
> csound"
>



-- 


Andrés


Send bugs reports to this list.
To unsubscribe, send email sympa@lists.bath.ac.uk with body "unsubscribe csound"

The convolution opcodes employ linear convolution (zero-padded fft 
blocks) in the complex domain; you will not get the same result using 
pvs-based multiplication which is "circular" convolution (I am not sure 
if PVS_COMPLEX is even supported yet in the pvs opcodes). I would also 
say that long IRs (define "long") in  stereo at 96KHz is genuinely quite 
a lot of processing (the fast convolution process itself involves a fair 
amount of memoery manipulation).  Ssing a really fast FFT library like 
FFTW would help of course, but we don't have that luxury being "merely" 
LGPL.


Richard Dobson

On 08/02/2010 16:13, Peiman Khosravi wrote:
> Hello,
>
> I've been playing around a little with the convolution opcodes. They
> seem very CPU demanding and at sr 96khz it is almost impossible to run
> long stereo IR files in real-time. I'm thinking of making a plug-in with
> csladspa for ardour. Has anyone got any experience with these opcodes?
> What is the most efficient and at the same time high quality opcode to
> use? Would it be better to simply multiply two pv-streams with the pvs
> opcodes?
>
> Thanks in advance.
>
> Best,
>
> Peiman
>
>
> Send bugs reports to this list.
> To unsubscribe, send email sympa@lists.bath.ac.uk with body "unsubscribe
> csound"
>




Send bugs reports to this list.
To unsubscribe, send email sympa@lists.bath.ac.uk with body "unsubscribe csound"

Hello,

Thanks for the reply. Yes I think you are right about pvs not doing  
the job here, I didn't think about that.

I am trying pvconvolve now and it sounds fine to me. Problem is that I  
get glitches even at higher ksmps. I think it's because of the 96khz  
sampling rate.

I'm attaching a test file that runs with QuteCsound + the IR. Maybe  
I'm doing something wrong?

Thanks!

Peiman


Send bugs reports to this list.
To unsubscribe, send email sympa@lists.bath.ac.uk with body "unsubscribe csound"


On 8 Feb 2010, at 16:23, Andres Cabrera wrote:

> Hi,
>
> Have you tried pconvolve and different values of ksmps?
>
> They should all give you the same quality, unless there's something
> wrong with them.
>
> I'm not sure using the pvs opcodes will work, as they are windowed. Or
> will they?
>
> Cheers,
> Andrés
>
> On Mon, Feb 8, 2010 at 4:13 PM, Peiman Khosravi
>  wrote:
>> Hello,
>>
>> I've been playing around a little with the convolution opcodes.  
>> They seem
>> very CPU demanding and at sr 96khz it is almost impossible to run  
>> long
>> stereo IR files in real-time. I'm thinking of making a plug-in with  
>> csladspa
>> for ardour. Has anyone got any experience with these opcodes? What  
>> is the
>> most efficient and at the same time high quality opcode to use?  
>> Would it be
>> better to simply multiply two pv-streams with the pvs opcodes?
>>
>> Thanks in advance.
>>
>> Best,
>>
>> Peiman
>>
>>
>> Send bugs reports to this list.
>> To unsubscribe, send email sympa@lists.bath.ac.uk with body  
>> "unsubscribe
>> csound"
>>
>
>
>
> -- 
>
>
> Andrés
>
>
> Send bugs reports to this list.
> To unsubscribe, send email sympa@lists.bath.ac.uk with body  
> "unsubscribe csound"

Thanks very much. When I say long I mean in the region of 3 seconds,  
so not extreme. I see your point about the sampling-rate. What would  
happen if the ir is not the same sr as the csd?

Best,

Peiman

On 8 Feb 2010, at 16:33, Richard Dobson wrote:

> The convolution opcodes employ linear convolution (zero-padded fft  
> blocks) in the complex domain; you will not get the same result  
> using pvs-based multiplication which is "circular" convolution (I am  
> not sure if PVS_COMPLEX is even supported yet in the pvs opcodes). I  
> would also say that long IRs (define "long") in  stereo at 96KHz is  
> genuinely quite a lot of processing (the fast convolution process  
> itself involves a fair amount of memoery manipulation).  Ssing a  
> really fast FFT library like FFTW would help of course, but we don't  
> have that luxury being "merely" LGPL.
>
>
> Richard Dobson
>
> On 08/02/2010 16:13, Peiman Khosravi wrote:
>> Hello,
>>
>> I've been playing around a little with the convolution opcodes. They
>> seem very CPU demanding and at sr 96khz it is almost impossible to  
>> run
>> long stereo IR files in real-time. I'm thinking of making a plug-in  
>> with
>> csladspa for ardour. Has anyone got any experience with these  
>> opcodes?
>> What is the most efficient and at the same time high quality opcode  
>> to
>> use? Would it be better to simply multiply two pv-streams with the  
>> pvs
>> opcodes?
>>
>> Thanks in advance.
>>
>> Best,
>>
>> Peiman
>>
>>
>> Send bugs reports to this list.
>> To unsubscribe, send email sympa@lists.bath.ac.uk with body  
>> "unsubscribe
>> csound"
>>
>
>
>
>
> Send bugs reports to this list.
> To unsubscribe, send email sympa@lists.bath.ac.uk with body  
> "unsubscribe csound"



Send bugs reports to this list.
To unsubscribe, send email sympa@lists.bath.ac.uk with body "unsubscribe csound"

I made two examples for QuteCsound, one with pconvolve ("Simple  
Convolution"), and one with ftconv ("Universal Convolution"). Perhaps  
you want to have a look. I found ftconv to be the most effective and  
flexible. At both opcodes the performance depends on the partition  
size. As Matt says in the manual page for pconvolve: "tweaking" ...
Ciao -

	joachim


Am 08.02.2010 um 17:13 schrieb Peiman Khosravi:

> Hello,
>
> I've been playing around a little with the convolution opcodes. They  
> seem very CPU demanding and at sr 96khz it is almost impossible to  
> run long stereo IR files in real-time. I'm thinking of making a plug- 
> in with csladspa for ardour. Has anyone got any experience with  
> these opcodes? What is the most efficient and at the same time high  
> quality opcode to use? Would it be better to simply multiply two pv- 
> streams with the pvs opcodes?
>
> Thanks in advance.
>
> Best,
>
> Peiman
>
>
> Send bugs reports to this list.
> To unsubscribe, send email sympa@lists.bath.ac.uk with body  
> "unsubscribe csound"
>



Send bugs reports to this list.
To unsubscribe, send email sympa@lists.bath.ac.uk with body "unsubscribe csound"

Thanks very much, I hadn't seen the examples so I'll have  look.

Best,

Peiman

On 8 Feb 2010, at 18:21, joachim heintz wrote:

> I made two examples for QuteCsound, one with pconvolve ("Simple  
> Convolution"), and one with ftconv ("Universal Convolution").  
> Perhaps you want to have a look. I found ftconv to be the most  
> effective and flexible. At both opcodes the performance depends on  
> the partition size. As Matt says in the manual page for pconvolve:  
> "tweaking" ...
> Ciao -
>
> 	joachim
>
>
> Am 08.02.2010 um 17:13 schrieb Peiman Khosravi:
>
>> Hello,
>>
>> I've been playing around a little with the convolution opcodes.  
>> They seem very CPU demanding and at sr 96khz it is almost  
>> impossible to run long stereo IR files in real-time. I'm thinking  
>> of making a plug-in with csladspa for ardour. Has anyone got any  
>> experience with these opcodes? What is the most efficient and at  
>> the same time high quality opcode to use? Would it be better to  
>> simply multiply two pv-streams with the pvs opcodes?
>>
>> Thanks in advance.
>>
>> Best,
>>
>> Peiman
>>
>>
>> Send bugs reports to this list.
>> To unsubscribe, send email sympa@lists.bath.ac.uk with body  
>> "unsubscribe csound"
>>
>
>
>
> Send bugs reports to this list.
> To unsubscribe, send email sympa@lists.bath.ac.uk with body  
> "unsubscribe csound"



Send bugs reports to this list.
To unsubscribe, send email sympa@lists.bath.ac.uk with body "unsubscribe csound"