[Csnd] detecting noisiness

[Csnd] detecting noisiness

Date	2013-11-24 12:44
From	peiman khosravi
Subject	[Csnd] detecting noisiness
	Is there an opcode for detecting signal periodicity? I want to know the noisiness of an input signal, regardless of the periodic content in might contain (e.g. how much breath tone there is in a flute tone). I know how to do this in maxmsp (there is a nice external by Tristan Jehan) but haven't got a pd solution yet so I'm wondering if csound offers something. Otherwise, how does one go about calculating noisiness from FFT data? Thanks Peiman www.peimankhosravi.co.uk \|\| RSS Feed \|\| Concert News

Date	2013-11-24 12:47
From	peiman khosravi
Subject	[Csnd] Re: detecting noisiness
	Upon further examination. The msp external uses "bark-based spectral flatness measure". But I don't know what that means! Thanks Peiman www.peimankhosravi.co.uk \|\| RSS Feed \|\| Concert News On 24 November 2013 12:44, peiman khosravi <peimankhosravi@gmail.com> wrote: Is there an opcode for detecting signal periodicity? I want to know the noisiness of an input signal, regardless of the periodic content in might contain (e.g. how much breath tone there is in a flute tone). I know how to do this in maxmsp (there is a nice external by Tristan Jehan) but haven't got a pd solution yet so I'm wondering if csound offers something. Otherwise, how does one go about calculating noisiness from FFT data? Thanks Peiman www.peimankhosravi.co.uk \|\| RSS Feed \|\| Concert News

Date	2013-11-24 13:52
From	Anders Genell
Subject	Re: [Csnd] Re: detecting noisiness
	I would assume it does some analysis of the deviation of each bark band from the average level of all bands. Bark bands are very similar to standard 1/3 octave bands. If only one bark band is excited the signal is likely to be tonal or narrow band noise. Problems arise when you have several adjacent tones falling into adjacent bands. Then it is hard to use bark band analysis for tonality. Regards, Anders On Sun, Nov 24, 2013 at 1:47 PM, peiman khosravi <peimankhosravi@gmail.com> wrote: Upon further examination. The msp external uses "bark-based spectral flatness measure". But I don't know what that means! Thanks Peiman www.peimankhosravi.co.uk \|\| RSS Feed \|\| Concert News On 24 November 2013 12:44, peiman khosravi <peimankhosravi@gmail.com> wrote: Is there an opcode for detecting signal periodicity? I want to know the noisiness of an input signal, regardless of the periodic content in might contain (e.g. how much breath tone there is in a flute tone). I know how to do this in maxmsp (there is a nice external by Tristan Jehan) but haven't got a pd solution yet so I'm wondering if csound offers something. Otherwise, how does one go about calculating noisiness from FFT data? Thanks Peiman www.peimankhosravi.co.uk \|\| RSS Feed \|\| Concert News

Date	2013-11-24 14:43
From	Joel Ross
Subject	Re: [Csnd] Re: detecting noisiness
	At a guess based on information here: http://en.wikipedia.org/wiki/Spectral_flatness and here: https://ccrma.stanford.edu/courses/120-fall-2003/lecture-5.html You would take the geometric mean of the magnitudes of the bark transformed spectrum and divide by the arithmetic mean of the same. In other words, take the magnitude of the fft and convert to bark scale before producing the two means and dividing. Regards, Joel On 24 November 2013 13:52, Anders Genell wrote: > I would assume it does some analysis of the deviation of each bark band from > the average level of all bands. Bark bands are very similar to standard 1/3 > octave bands. > If only one bark band is excited the signal is likely to be tonal or narrow > band noise. Problems arise when you have several adjacent tones falling into > adjacent bands. Then it is hard to use bark band analysis for tonality. > > Regards, > Anders > > > On Sun, Nov 24, 2013 at 1:47 PM, peiman khosravi > wrote: >> >> Upon further examination. The msp external uses "bark-based spectral >> flatness measure". But I don't know what that means! >> >> Thanks >> Peiman >> >> >> >> >> www.peimankhosravi.co.uk \|\| RSS Feed \|\| Concert News >> >> >> On 24 November 2013 12:44, peiman khosravi >> wrote: >>> >>> Is there an opcode for detecting signal periodicity? I want to know the >>> noisiness of an input signal, regardless of the periodic content in might >>> contain (e.g. how much breath tone there is in a flute tone). I know how to >>> do this in maxmsp (there is a nice external by Tristan Jehan) but haven't >>> got a pd solution yet so I'm wondering if csound offers something. >>> Otherwise, how does one go about calculating noisiness from FFT data? >>> >>> Thanks >>> Peiman >>> >>> >>> >>> www.peimankhosravi.co.uk \|\| RSS Feed \|\| Concert News >> >> >

Date	2013-11-24 14:51
From	Justin Smith
Subject	Re: [Csnd] Re: detecting noisiness
	If you want noisiness as a human listener would experience it, you should be measuring critical band overlap. The critical band changes in width throughout the spectrum. http://en.wikipedia.org/wiki/Critical_band The critical band expands at a greater rate than frequency (such that, overall the same interval is more dissonant at higher pitches). On Sun, Nov 24, 2013 at 6:43 AM, Joel Ross <joel.binarybrain@gmail.com> wrote: At a guess based on information here: http://en.wikipedia.org/wiki/Spectral_flatness and here: https://ccrma.stanford.edu/courses/120-fall-2003/lecture-5.html You would take the geometric mean of the magnitudes of the bark transformed spectrum and divide by the arithmetic mean of the same. In other words, take the magnitude of the fft and convert to bark scale before producing the two means and dividing. Regards, Joel On 24 November 2013 13:52, Anders Genell <anders.genell@gmail.com> wrote: > I would assume it does some analysis of the deviation of each bark band from > the average level of all bands. Bark bands are very similar to standard 1/3 > octave bands. > If only one bark band is excited the signal is likely to be tonal or narrow > band noise. Problems arise when you have several adjacent tones falling into > adjacent bands. Then it is hard to use bark band analysis for tonality. > > Regards, > Anders > > > On Sun, Nov 24, 2013 at 1:47 PM, peiman khosravi <peimankhosravi@gmail.com> > wrote: >> >> Upon further examination. The msp external uses "bark-based spectral >> flatness measure". But I don't know what that means! >> >> Thanks >> Peiman >> >> >> >> >> www.peimankhosravi.co.uk \|\| RSS Feed \|\| Concert News >> >> >> On 24 November 2013 12:44, peiman khosravi <peimankhosravi@gmail.com> >> wrote: >>> >>> Is there an opcode for detecting signal periodicity? I want to know the >>> noisiness of an input signal, regardless of the periodic content in might >>> contain (e.g. how much breath tone there is in a flute tone). I know how to >>> do this in maxmsp (there is a nice external by Tristan Jehan) but haven't >>> got a pd solution yet so I'm wondering if csound offers something. >>> Otherwise, how does one go about calculating noisiness from FFT data? >>> >>> Thanks >>> Peiman >>> >>> >>> >>> www.peimankhosravi.co.uk \|\| RSS Feed \|\| Concert News >> >> > Send bugs reports to the Sourceforge bug trackers csound6: https://sourceforge.net/p/csound/tickets/ csound5: https://sourceforge.net/p/csound/bugs/ Discussions of bugs and features can be posted here To unsubscribe, send email sympa@lists.bath.ac.uk with body "unsubscribe csound"

Date	2013-11-24 15:21
From	Joel Ross
Subject	Re: [Csnd] Re: detecting noisiness
	The bark scale is intended to account for the critical band in the method above. However, only examining the critical band would give you a measure of roughness rather than noisiness. What the spectral flatness is doing is estimating how close to the spectral density of white noise the signal is or in other words, how 'flat' it is. This is a statistical measurement and it makes assumptions about the thing you are looking at. You could trick the process by having a periodic sound with a broadly flat spectrum. It should be reasonable for a range of real world sounds however. Joel On 24 November 2013 14:51, Justin Smith wrote: > If you want noisiness as a human listener would experience it, you should be > measuring critical band overlap. The critical band changes in width > throughout the spectrum. > > http://en.wikipedia.org/wiki/Critical_band > > The critical band expands at a greater rate than frequency (such that, > overall the same interval is more dissonant at higher pitches). > > > On Sun, Nov 24, 2013 at 6:43 AM, Joel Ross > wrote: >> >> At a guess based on information here: >> http://en.wikipedia.org/wiki/Spectral_flatness >> and here: >> https://ccrma.stanford.edu/courses/120-fall-2003/lecture-5.html >> >> You would take the geometric mean of the magnitudes of the bark >> transformed spectrum and divide by the arithmetic mean of the same. In >> other words, take the magnitude of the fft and convert to bark scale >> before producing the two means and dividing. >> >> Regards, >> Joel >> >> On 24 November 2013 13:52, Anders Genell wrote: >> > I would assume it does some analysis of the deviation of each bark band >> > from >> > the average level of all bands. Bark bands are very similar to standard >> > 1/3 >> > octave bands. >> > If only one bark band is excited the signal is likely to be tonal or >> > narrow >> > band noise. Problems arise when you have several adjacent tones falling >> > into >> > adjacent bands. Then it is hard to use bark band analysis for tonality. >> > >> > Regards, >> > Anders >> > >> > >> > On Sun, Nov 24, 2013 at 1:47 PM, peiman khosravi >> > >> > wrote: >> >> >> >> Upon further examination. The msp external uses "bark-based spectral >> >> flatness measure". But I don't know what that means! >> >> >> >> Thanks >> >> Peiman >> >> >> >> >> >> >> >> >> >> www.peimankhosravi.co.uk \|\| RSS Feed \|\| Concert News >> >> >> >> >> >> On 24 November 2013 12:44, peiman khosravi >> >> wrote: >> >>> >> >>> Is there an opcode for detecting signal periodicity? I want to know >> >>> the >> >>> noisiness of an input signal, regardless of the periodic content in >> >>> might >> >>> contain (e.g. how much breath tone there is in a flute tone). I know >> >>> how to >> >>> do this in maxmsp (there is a nice external by Tristan Jehan) but >> >>> haven't >> >>> got a pd solution yet so I'm wondering if csound offers something. >> >>> Otherwise, how does one go about calculating noisiness from FFT data? >> >>> >> >>> Thanks >> >>> Peiman >> >>> >> >>> >> >>> >> >>> www.peimankhosravi.co.uk \|\| RSS Feed \|\| Concert News >> >> >> >> >> > >> >> >> Send bugs reports to the Sourceforge bug trackers >> csound6: >> https://sourceforge.net/p/csound/tickets/ >> csound5: >> https://sourceforge.net/p/csound/bugs/ >> Discussions of bugs and features can be posted here >> To unsubscribe, send email sympa@lists.bath.ac.uk with body "unsubscribe >> csound" >> >> >

Date	2013-11-24 15:38
From	peiman khosravi
Subject	Re: [Csnd] Re: detecting noisiness
	Hello, Thanks for the replies. All makes perfect sense. So I'm not so concerned about roughness. I just need to calculate the periodicity of a violin sound. So I gather than if I use a combination of flatness and flux (how much does each bin change from window to window) then I should be able to get a good estimate right? In this case, a periodic signal with a flat spectral energy distribution would not trick the process? Thanks, Peiman www.peimankhosravi.co.uk \|\| RSS Feed \|\| Concert News On 24 November 2013 15:21, Joel Ross <joel.binarybrain@gmail.com> wrote: The bark scale is intended to account for the critical band in the method above. However, only examining the critical band would give you a measure of roughness rather than noisiness. What the spectral flatness is doing is estimating how close to the spectral density of white noise the signal is or in other words, how 'flat' it is. This is a statistical measurement and it makes assumptions about the thing you are looking at. You could trick the process by having a periodic sound with a broadly flat spectrum. It should be reasonable for a range of real world sounds however. Joel On 24 November 2013 14:51, Justin Smith <noisesmith@gmail.com> wrote: > If you want noisiness as a human listener would experience it, you should be > measuring critical band overlap. The critical band changes in width > throughout the spectrum. > > http://en.wikipedia.org/wiki/Critical_band > > The critical band expands at a greater rate than frequency (such that, > overall the same interval is more dissonant at higher pitches). > > > On Sun, Nov 24, 2013 at 6:43 AM, Joel Ross <joel.binarybrain@gmail.com> > wrote: >> >> At a guess based on information here: >> http://en.wikipedia.org/wiki/Spectral_flatness >> and here: >> https://ccrma.stanford.edu/courses/120-fall-2003/lecture-5.html >> >> You would take the geometric mean of the magnitudes of the bark >> transformed spectrum and divide by the arithmetic mean of the same. In >> other words, take the magnitude of the fft and convert to bark scale >> before producing the two means and dividing. >> >> Regards, >> Joel >> >> On 24 November 2013 13:52, Anders Genell <anders.genell@gmail.com> wrote: >> > I would assume it does some analysis of the deviation of each bark band >> > from >> > the average level of all bands. Bark bands are very similar to standard >> > 1/3 >> > octave bands. >> > If only one bark band is excited the signal is likely to be tonal or >> > narrow >> > band noise. Problems arise when you have several adjacent tones falling >> > into >> > adjacent bands. Then it is hard to use bark band analysis for tonality. >> > >> > Regards, >> > Anders >> > >> > >> > On Sun, Nov 24, 2013 at 1:47 PM, peiman khosravi >> > <peimankhosravi@gmail.com> >> > wrote: >> >> >> >> Upon further examination. The msp external uses "bark-based spectral >> >> flatness measure". But I don't know what that means! >> >> >> >> Thanks >> >> Peiman >> >> >> >> >> >> >> >> >> >> www.peimankhosravi.co.uk \|\| RSS Feed \|\| Concert News >> >> >> >> >> >> On 24 November 2013 12:44, peiman khosravi <peimankhosravi@gmail.com> >> >> wrote: >> >>> >> >>> Is there an opcode for detecting signal periodicity? I want to know >> >>> the >> >>> noisiness of an input signal, regardless of the periodic content in >> >>> might >> >>> contain (e.g. how much breath tone there is in a flute tone). I know >> >>> how to >> >>> do this in maxmsp (there is a nice external by Tristan Jehan) but >> >>> haven't >> >>> got a pd solution yet so I'm wondering if csound offers something. >> >>> Otherwise, how does one go about calculating noisiness from FFT data? >> >>> >> >>> Thanks >> >>> Peiman >> >>> >> >>> >> >>> >> >>> www.peimankhosravi.co.uk \|\| RSS Feed \|\| Concert News >> >> >> >> >> > >> >> >> Send bugs reports to the Sourceforge bug trackers >> csound6: >> https://sourceforge.net/p/csound/tickets/ >> csound5: >> https://sourceforge.net/p/csound/bugs/ >> 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 trackers csound6: https://sourceforge.net/p/csound/tickets/ csound5: https://sourceforge.net/p/csound/bugs/ Discussions of bugs and features can be posted here To unsubscribe, send email sympa@lists.bath.ac.uk with body "unsubscribe csound"