[Csnd] csound as a DSP

[Csnd] csound as a DSP

Date	2012-03-06 17:36
From	Peter Schneider-Kamp
Subject	[Csnd] csound as a DSP
	Dear all, I am new to the list (and to csound). First a big thanks for a great swiss army knife of sound! While I hold a personal interest in sound generation and have written a small additive synthesiser in Python for fun, my main interest in csound is in using it as a digital signal processor. Right now, I am using instruments based on pconvolve for digital room correction as well as for reverbation effects. So basically I use it as something where 2 audio channels enter, get processed, and (with a small delay) exit again in realtime. Now my question is whether I am doing things in the right way: a) Is the performance ok like this? Is there a better solution? b) How can I get rid of the ugly 3600 seconds and run csound indefinitely? c) Is there a way to save CPU power by detecting an "empty" (all zero) input? -odac3 -iadc2 ; from Soundflower to Focusrite USB interface sr = 44100 ksmps = 32 nchnls = 2 0dbfs = 1.0 instr 1 al, ar ins al1 pconvolve al, "fir_left.wav", p4 ar1 pconvolve ar, "fir_right.wav", p4 outs 0.9al1, 0.9ar1 endin i 1 0 3600 512 ; partition size as p4 e Cheers, Peter -- Peter Schneider-Kamp mailto:petersk@imada.sdu.dk Math. & Computer Science http://imada.sdu.dk/~petersk/ University of Southern Denmark phone +45 6550 2327

Date	2012-03-06 17:54
From	Victor Lazzarini
Subject	Re: [Csnd] csound as a DSP
	If you need low latency, then try to get your buffer size as low as possible, with -b . If you don't set the partitionsize, then this (according to the manual) will also set your partitionsize. A suggestion would be to use -b 32 or -b 64 if you can get it that low. Victor On 6 Mar 2012, at 17:36, Peter Schneider-Kamp wrote: > Dear all, > > I am new to the list (and to csound). First a big thanks for a great > swiss army knife of sound! > > While I hold a personal interest in sound generation and have written > a small additive synthesiser in Python for fun, my main interest in > csound is in using it as a digital signal processor. > > Right now, I am using instruments based on pconvolve for digital > room correction as well as for reverbation effects. So basically I > use it as something where 2 audio channels enter, get processed, and > (with a small delay) exit again in realtime. > > Now my question is whether I am doing things in the right way: > > a) Is the performance ok like this? Is there a better solution? > b) How can I get rid of the ugly 3600 seconds and run csound > indefinitely? > c) Is there a way to save CPU power by detecting an "empty" > (all zero) input? > > > > -odac3 -iadc2 ; from Soundflower to Focusrite USB interface > > > sr = 44100 > ksmps = 32 > nchnls = 2 > 0dbfs = 1.0 > > instr 1 > al, ar ins > al1 pconvolve al, "fir_left.wav", p4 > ar1 pconvolve ar, "fir_right.wav", p4 > outs 0.9al1, 0.9ar1 > endin > > > i 1 0 3600 512 ; partition size as p4 > e > > > > Cheers, > Peter > > -- > Peter Schneider-Kamp mailto:petersk@imada.sdu.dk > Math. & Computer Science http://imada.sdu.dk/~petersk/ > University of Southern Denmark phone +45 6550 2327 > > > Send bugs reports to the Sourceforge bug tracker > https://sourceforge.net/tracker/?group_id=81968&atid=564599 > Discussions of bugs and features can be posted here > To unsubscribe, send email sympa@lists.bath.ac.uk with body "unsubscribe csound" > Dr Victor Lazzarini Senior Lecturer Dept. of Music NUI Maynooth Ireland tel.: +353 1 708 3545 Victor dot Lazzarini AT nuim dot ie

Date	2012-03-06 19:07
From	Peter Schneider-Kamp
Subject	Re: [Csnd] csound as a DSP
	Dear Victor, thanks for suggesting tuning of buffer sizes. I had not yet thought about it. I will perform some experiments with that :) I do have adjusted the partitionsize for a relatively low latency but low CPU setting. Here are my measurements on a 2.66 GHz Core i 7 Macbook Pro 15" (CPU load in percentage and length of FIR in seconds): Part.sizee Delay CPU @ 0.5s CPU @ 5.3s 8 0.000181 98.4 16 0.000363 54.1 32 0.000726 27.6 64 0.002177 16.2 128 0.003628 9.4 99.3 256 0.006531 5.4 53.7 512 0.012336 3.5 28.6 1024 0.023946 2.5 15.3 2048 0.047166 2.2 8.2 4096 0.093605 1.8 4.9 8192 0.186485 1.6 3.1 16384 0.372245 2.4 32768 0.743764 2.2 As I am planning to use the room correction on a Raspberry Pi (700 MHz ARM processor, 224 MB RAM), I guess that I cannot go below 2048 (i.e. 47ms delay) for a 5s FIR or below 128 (i.e. 3.6ms delay) for a 0.5s FIR. Basically, I guess I need to make sure that I get a good averaged FIR from a number of measurements with at most 3s length in order to push the delay down to less than 20ms, i.e., to something that will not bother me for video synchronicity. Cheers, Peter On 3/6/12 6:54 PM, Victor Lazzarini wrote: > If you need low latency, then try to get your buffer size as low as possible, with -b . If you don't set the partitionsize, then this (according to the manual) will also set your partitionsize. > A suggestion would be to use -b 32 or -b 64 if you can get it that low. > > Victor > > On 6 Mar 2012, at 17:36, Peter Schneider-Kamp wrote: > >> Dear all, >> >> I am new to the list (and to csound). First a big thanks for a great >> swiss army knife of sound! >> >> While I hold a personal interest in sound generation and have written >> a small additive synthesiser in Python for fun, my main interest in >> csound is in using it as a digital signal processor. >> >> Right now, I am using instruments based on pconvolve for digital >> room correction as well as for reverbation effects. So basically I >> use it as something where 2 audio channels enter, get processed, and >> (with a small delay) exit again in realtime. >> >> Now my question is whether I am doing things in the right way: >> >> a) Is the performance ok like this? Is there a better solution? >> b) How can I get rid of the ugly 3600 seconds and run csound >> indefinitely? >> c) Is there a way to save CPU power by detecting an "empty" >> (all zero) input? >> >> >> >> -odac3 -iadc2 ; from Soundflower to Focusrite USB interface >> >> >> sr = 44100 >> ksmps = 32 >> nchnls = 2 >> 0dbfs = 1.0 >> >> instr 1 >> al, ar ins >> al1 pconvolve al, "fir_left.wav", p4 >> ar1 pconvolve ar, "fir_right.wav", p4 >> outs 0.9al1, 0.9ar1 >> endin >> >> >> i 1 0 3600 512 ; partition size as p4 >> e >> >> >> >> Cheers, >> Peter >> >> -- >> Peter Schneider-Kamp mailto:petersk@imada.sdu.dk >> Math.& Computer Science http://imada.sdu.dk/~petersk/ >> University of Southern Denmark phone +45 6550 2327 >> >> >> Send bugs reports to the Sourceforge bug tracker >> https://sourceforge.net/tracker/?group_id=81968&atid=564599 >> Discussions of bugs and features can be posted here >> To unsubscribe, send email sympa@lists.bath.ac.uk with body "unsubscribe csound" >> > > Dr Victor Lazzarini > Senior Lecturer > Dept. of Music > NUI Maynooth Ireland > tel.: +353 1 708 3545 > Victor dot Lazzarini AT nuim dot ie > > > > > > Send bugs reports to the Sourceforge bug tracker > https://sourceforge.net/tracker/?group_id=81968&atid=564599 > Discussions of bugs and features can be posted here > To unsubscribe, send email sympa@lists.bath.ac.uk with body "unsubscribe csound" > -- Peter Schneider-Kamp mailto:petersk@imada.sdu.dk Math. & Computer Science http://imada.sdu.dk/~petersk/ University of Southern Denmark phone +45 6550 2327

Date	2012-03-06 21:40
From	joachim heintz
Subject	Re: [Csnd] csound as a DSP
	hi peter - as to the performance, you will save a lot of cpu if you increase your ksmps. depending on the partition size you choose, this may perhaps not introduce more latency. you can try ksmps=64 / -b 128 / -B 128, or even 128 / 256 / 256. best - joachim Am 06.03.2012 20:07, schrieb Peter Schneider-Kamp: > Dear Victor, > > thanks for suggesting tuning of buffer sizes. I had not yet thought > about it. I will perform some experiments with that :) > > I do have adjusted the partitionsize for a relatively low > latency but low CPU setting. Here are my measurements on a > 2.66 GHz Core i 7 Macbook Pro 15" (CPU load in percentage > and length of FIR in seconds): > > Part.sizee Delay CPU @ 0.5s CPU @ 5.3s > 8 0.000181 98.4 > 16 0.000363 54.1 > 32 0.000726 27.6 > 64 0.002177 16.2 > 128 0.003628 9.4 99.3 > 256 0.006531 5.4 53.7 > 512 0.012336 3.5 28.6 > 1024 0.023946 2.5 15.3 > 2048 0.047166 2.2 8.2 > 4096 0.093605 1.8 4.9 > 8192 0.186485 1.6 3.1 > 16384 0.372245 2.4 > 32768 0.743764 2.2 > > As I am planning to use the room correction on a Raspberry Pi > (700 MHz ARM processor, 224 MB RAM), I guess that I cannot > go below 2048 (i.e. 47ms delay) for a 5s FIR or below 128 > (i.e. 3.6ms delay) for a 0.5s FIR. > > Basically, I guess I need to make sure that I get a good averaged > FIR from a number of measurements with at most 3s length in order > to push the delay down to less than 20ms, i.e., to something that will > not bother me for video synchronicity. > > Cheers, > Peter > > On 3/6/12 6:54 PM, Victor Lazzarini wrote: >> If you need low latency, then try to get your buffer size as low as >> possible, with -b . If you don't set the partitionsize, then >> this (according to the manual) will also set your partitionsize. >> A suggestion would be to use -b 32 or -b 64 if you can get it that low. >> >> Victor >> >> On 6 Mar 2012, at 17:36, Peter Schneider-Kamp wrote: >> >>> Dear all, >>> >>> I am new to the list (and to csound). First a big thanks for a great >>> swiss army knife of sound! >>> >>> While I hold a personal interest in sound generation and have written >>> a small additive synthesiser in Python for fun, my main interest in >>> csound is in using it as a digital signal processor. >>> >>> Right now, I am using instruments based on pconvolve for digital >>> room correction as well as for reverbation effects. So basically I >>> use it as something where 2 audio channels enter, get processed, and >>> (with a small delay) exit again in realtime. >>> >>> Now my question is whether I am doing things in the right way: >>> >>> a) Is the performance ok like this? Is there a better solution? >>> b) How can I get rid of the ugly 3600 seconds and run csound >>> indefinitely? >>> c) Is there a way to save CPU power by detecting an "empty" >>> (all zero) input? >>> >>> >>> >>> -odac3 -iadc2 ; from Soundflower to Focusrite USB interface >>> >>> >>> sr = 44100 >>> ksmps = 32 >>> nchnls = 2 >>> 0dbfs = 1.0 >>> >>> instr 1 >>> al, ar ins >>> al1 pconvolve al, "fir_left.wav", p4 >>> ar1 pconvolve ar, "fir_right.wav", p4 >>> outs 0.9al1, 0.9ar1 >>> endin >>> >>> >>> i 1 0 3600 512 ; partition size as p4 >>> e >>> >>> >>> >>> Cheers, >>> Peter >>> >>> -- >>> Peter Schneider-Kamp mailto:petersk@imada.sdu.dk >>> Math.& Computer Science http://imada.sdu.dk/~petersk/ >>> University of Southern Denmark phone +45 6550 2327 >>> >>> >>> Send bugs reports to the Sourceforge bug tracker >>> https://sourceforge.net/tracker/?group_id=81968&atid=564599 >>> Discussions of bugs and features can be posted here >>> To unsubscribe, send email sympa@lists.bath.ac.uk with body >>> "unsubscribe csound" >>> >> >> Dr Victor Lazzarini >> Senior Lecturer >> Dept. of Music >> NUI Maynooth Ireland >> tel.: +353 1 708 3545 >> Victor dot Lazzarini AT nuim dot ie >> >> >> >> >> >> Send bugs reports to the Sourceforge bug tracker >> https://sourceforge.net/tracker/?group_id=81968&atid=564599 >> Discussions of bugs and features can be posted here >> To unsubscribe, send email sympa@lists.bath.ac.uk with body >> "unsubscribe csound" >> > >

Date	2012-03-07 16:39
From	Kelly Hirai
Subject	Re: [Csnd] csound as a DSP
	i have been thinking about using the pi for sound. i'm not seeing a way to easily bring low latency audio in. there is a SPI bus but i don't know if that is suitable. the docs for the bcm2835 suggest there are stereo i2s: rx tx. i assume the tx is prewired for the 1/8 phone jack, but i cant find any analog line in on this board or pinouts that suggest the i2s rx is available. nor am i sure of how to wire a ADC to it if i could find it. any hints? kelly On 03/06/12 16:40, joachim heintz wrote: > hi peter - > as to the performance, you will save a lot of cpu if you increase your > ksmps. > depending on the partition size you choose, this may perhaps not > introduce more latency. > you can try ksmps=64 / -b 128 / -B 128, or even 128 / 256 / 256. > best - > joachim > > > Am 06.03.2012 20:07, schrieb Peter Schneider-Kamp: >> Dear Victor, >> >> thanks for suggesting tuning of buffer sizes. I had not yet thought >> about it. I will perform some experiments with that :) >> >> I do have adjusted the partitionsize for a relatively low >> latency but low CPU setting. Here are my measurements on a >> 2.66 GHz Core i 7 Macbook Pro 15" (CPU load in percentage >> and length of FIR in seconds): >> >> Part.sizee Delay CPU @ 0.5s CPU @ 5.3s >> 8 0.000181 98.4 >> 16 0.000363 54.1 >> 32 0.000726 27.6 >> 64 0.002177 16.2 >> 128 0.003628 9.4 99.3 >> 256 0.006531 5.4 53.7 >> 512 0.012336 3.5 28.6 >> 1024 0.023946 2.5 15.3 >> 2048 0.047166 2.2 8.2 >> 4096 0.093605 1.8 4.9 >> 8192 0.186485 1.6 3.1 >> 16384 0.372245 2.4 >> 32768 0.743764 2.2 >> >> As I am planning to use the room correction on a Raspberry Pi >> (700 MHz ARM processor, 224 MB RAM), I guess that I cannot >> go below 2048 (i.e. 47ms delay) for a 5s FIR or below 128 >> (i.e. 3.6ms delay) for a 0.5s FIR. >> >> Basically, I guess I need to make sure that I get a good averaged >> FIR from a number of measurements with at most 3s length in order >> to push the delay down to less than 20ms, i.e., to something that will >> not bother me for video synchronicity. >> >> Cheers, >> Peter >> >> On 3/6/12 6:54 PM, Victor Lazzarini wrote: >>> If you need low latency, then try to get your buffer size as low as >>> possible, with -b . If you don't set the partitionsize, then >>> this (according to the manual) will also set your partitionsize. >>> A suggestion would be to use -b 32 or -b 64 if you can get it that low. >>> >>> Victor >>> >>> On 6 Mar 2012, at 17:36, Peter Schneider-Kamp wrote: >>> >>>> Dear all, >>>> >>>> I am new to the list (and to csound). First a big thanks for a great >>>> swiss army knife of sound! >>>> >>>> While I hold a personal interest in sound generation and have written >>>> a small additive synthesiser in Python for fun, my main interest in >>>> csound is in using it as a digital signal processor. >>>> >>>> Right now, I am using instruments based on pconvolve for digital >>>> room correction as well as for reverbation effects. So basically I >>>> use it as something where 2 audio channels enter, get processed, and >>>> (with a small delay) exit again in realtime. >>>> >>>> Now my question is whether I am doing things in the right way: >>>> >>>> a) Is the performance ok like this? Is there a better solution? >>>> b) How can I get rid of the ugly 3600 seconds and run csound >>>> indefinitely? >>>> c) Is there a way to save CPU power by detecting an "empty" >>>> (all zero) input? >>>> >>>> >>>> >>>> -odac3 -iadc2 ; from Soundflower to Focusrite USB interface >>>> >>>> >>>> sr = 44100 >>>> ksmps = 32 >>>> nchnls = 2 >>>> 0dbfs = 1.0 >>>> >>>> instr 1 >>>> al, ar ins >>>> al1 pconvolve al, "fir_left.wav", p4 >>>> ar1 pconvolve ar, "fir_right.wav", p4 >>>> outs 0.9al1, 0.9ar1 >>>> endin >>>> >>>> >>>> i 1 0 3600 512 ; partition size as p4 >>>> e >>>> >>>> >>>> >>>> Cheers, >>>> Peter >>>> >>>> -- >>>> Peter Schneider-Kamp mailto:petersk@imada.sdu.dk >>>> Math.& Computer Science http://imada.sdu.dk/~petersk/ >>>> University of Southern Denmark phone +45 6550 2327 >>>> >>>> >>>> Send bugs reports to the Sourceforge bug tracker >>>> https://sourceforge.net/tracker/?group_id=81968&atid=564599 >>>> Discussions of bugs and features can be posted here >>>> To unsubscribe, send email sympa@lists.bath.ac.uk with body >>>> "unsubscribe csound" >>>> >>> Dr Victor Lazzarini >>> Senior Lecturer >>> Dept. of Music >>> NUI Maynooth Ireland >>> tel.: +353 1 708 3545 >>> Victor dot Lazzarini AT nuim dot ie >>> >>> >>> >>> >>> >>> Send bugs reports to the Sourceforge bug tracker >>> https://sourceforge.net/tracker/?group_id=81968&atid=564599 >>> Discussions of bugs and features can be posted here >>> To unsubscribe, send email sympa@lists.bath.ac.uk with body >>> "unsubscribe csound" >>> >> > > Send bugs reports to the Sourceforge bug tracker > https://sourceforge.net/tracker/?group_id=81968&atid=564599 > Discussions of bugs and features can be posted here > To unsubscribe, send email sympa@lists.bath.ac.uk with body "unsubscribe csound" >

Date	2012-03-07 17:35
From	Victor
Subject	Re: [Csnd] csound as a DSP
	apparently there is no input in the codec, but of course, if there is alsa, there is the possibility of usb audio. victor On 7 Mar 2012, at 16:39, Kelly Hirai wrote: > i have been thinking about using the pi for sound. i'm not seeing a way > to easily bring low latency audio in. there is a SPI bus but i don't > know if that is suitable. the docs for the bcm2835 suggest there are > stereo i2s: rx tx. i assume the tx is prewired for the 1/8 phone jack, > but i cant find any analog line in on this board or pinouts that suggest > the i2s rx is available. nor am i sure of how to wire a ADC to it if i > could find it. any hints? > > kelly > > On 03/06/12 16:40, joachim heintz wrote: >> hi peter - >> as to the performance, you will save a lot of cpu if you increase your >> ksmps. >> depending on the partition size you choose, this may perhaps not >> introduce more latency. >> you can try ksmps=64 / -b 128 / -B 128, or even 128 / 256 / 256. >> best - >> joachim >> >> >> Am 06.03.2012 20:07, schrieb Peter Schneider-Kamp: >>> Dear Victor, >>> >>> thanks for suggesting tuning of buffer sizes. I had not yet thought >>> about it. I will perform some experiments with that :) >>> >>> I do have adjusted the partitionsize for a relatively low >>> latency but low CPU setting. Here are my measurements on a >>> 2.66 GHz Core i 7 Macbook Pro 15" (CPU load in percentage >>> and length of FIR in seconds): >>> >>> Part.sizee Delay CPU @ 0.5s CPU @ 5.3s >>> 8 0.000181 98.4 >>> 16 0.000363 54.1 >>> 32 0.000726 27.6 >>> 64 0.002177 16.2 >>> 128 0.003628 9.4 99.3 >>> 256 0.006531 5.4 53.7 >>> 512 0.012336 3.5 28.6 >>> 1024 0.023946 2.5 15.3 >>> 2048 0.047166 2.2 8.2 >>> 4096 0.093605 1.8 4.9 >>> 8192 0.186485 1.6 3.1 >>> 16384 0.372245 2.4 >>> 32768 0.743764 2.2 >>> >>> As I am planning to use the room correction on a Raspberry Pi >>> (700 MHz ARM processor, 224 MB RAM), I guess that I cannot >>> go below 2048 (i.e. 47ms delay) for a 5s FIR or below 128 >>> (i.e. 3.6ms delay) for a 0.5s FIR. >>> >>> Basically, I guess I need to make sure that I get a good averaged >>> FIR from a number of measurements with at most 3s length in order >>> to push the delay down to less than 20ms, i.e., to something that will >>> not bother me for video synchronicity. >>> >>> Cheers, >>> Peter >>> >>> On 3/6/12 6:54 PM, Victor Lazzarini wrote: >>>> If you need low latency, then try to get your buffer size as low as >>>> possible, with -b . If you don't set the partitionsize, then >>>> this (according to the manual) will also set your partitionsize. >>>> A suggestion would be to use -b 32 or -b 64 if you can get it that low. >>>> >>>> Victor >>>> >>>> On 6 Mar 2012, at 17:36, Peter Schneider-Kamp wrote: >>>> >>>>> Dear all, >>>>> >>>>> I am new to the list (and to csound). First a big thanks for a great >>>>> swiss army knife of sound! >>>>> >>>>> While I hold a personal interest in sound generation and have written >>>>> a small additive synthesiser in Python for fun, my main interest in >>>>> csound is in using it as a digital signal processor. >>>>> >>>>> Right now, I am using instruments based on pconvolve for digital >>>>> room correction as well as for reverbation effects. So basically I >>>>> use it as something where 2 audio channels enter, get processed, and >>>>> (with a small delay) exit again in realtime. >>>>> >>>>> Now my question is whether I am doing things in the right way: >>>>> >>>>> a) Is the performance ok like this? Is there a better solution? >>>>> b) How can I get rid of the ugly 3600 seconds and run csound >>>>> indefinitely? >>>>> c) Is there a way to save CPU power by detecting an "empty" >>>>> (all zero) input? >>>>> >>>>> >>>>> >>>>> -odac3 -iadc2 ; from Soundflower to Focusrite USB interface >>>>> >>>>> >>>>> sr = 44100 >>>>> ksmps = 32 >>>>> nchnls = 2 >>>>> 0dbfs = 1.0 >>>>> >>>>> instr 1 >>>>> al, ar ins >>>>> al1 pconvolve al, "fir_left.wav", p4 >>>>> ar1 pconvolve ar, "fir_right.wav", p4 >>>>> outs 0.9al1, 0.9ar1 >>>>> endin >>>>> >>>>> >>>>> i 1 0 3600 512 ; partition size as p4 >>>>> e >>>>> >>>>> >>>>> >>>>> Cheers, >>>>> Peter >>>>> >>>>> -- >>>>> Peter Schneider-Kamp mailto:petersk@imada.sdu.dk >>>>> Math.& Computer Science http://imada.sdu.dk/~petersk/ >>>>> University of Southern Denmark phone +45 6550 2327 >>>>> >>>>> >>>>> Send bugs reports to the Sourceforge bug tracker >>>>> https://sourceforge.net/tracker/?group_id=81968&atid=564599 >>>>> Discussions of bugs and features can be posted here >>>>> To unsubscribe, send email sympa@lists.bath.ac.uk with body >>>>> "unsubscribe csound" >>>>> >>>> Dr Victor Lazzarini >>>> Senior Lecturer >>>> Dept. of Music >>>> NUI Maynooth Ireland >>>> tel.: +353 1 708 3545 >>>> Victor dot Lazzarini AT nuim dot ie >>>> >>>> >>>> >>>> >>>> >>>> Send bugs reports to the Sourceforge bug tracker >>>> https://sourceforge.net/tracker/?group_id=81968&atid=564599 >>>> Discussions of bugs and features can be posted here >>>> To unsubscribe, send email sympa@lists.bath.ac.uk with body >>>> "unsubscribe csound" >>>> >>> >> >> Send bugs reports to the Sourceforge bug tracker >> https://sourceforge.net/tracker/?group_id=81968&atid=564599 >> Discussions of bugs and features can be posted here >> To unsubscribe, send email sympa@lists.bath.ac.uk with body "unsubscribe csound" >> > > > Send bugs reports to the Sourceforge bug tracker > https://sourceforge.net/tracker/?group_id=81968&atid=564599 > Discussions of bugs and features can be posted here > To unsubscribe, send email sympa@lists.bath.ac.uk with body "unsubscribe csound" >