fft guru needed

Sep 7, 2007 at 1:50pm

fft guru needed

it`s dawned on me that fft (pfft) shifting in max
sounds like crap, excuse me for the vocals.
there always seems to be a framesize drawback
that i do not accept as proper sound.
but altiverb for instance sounds quite accurate
and doesnt have the mpeg like distortion to it.

are there optimal settings for chunksize and overlap
or.. am i just stuck with a poor sound with pfft

#33580
Sep 7, 2007 at 2:53pm

if you havent, check out fft ease:

http://www.sarc.qub.ac.uk/~elyon/LyonSoftware/MaxMSP/FFTease/

should be of some help.

j

#112020
Sep 7, 2007 at 3:13pm

I’ve always found fft stuff to sound pretty bad in max also. So the
best thing I have found so far is the csound external and/or the
rtcmix~ external.

Peiman

On 7 Sep 2007, at 14:50, lukas wrote:

>
> it`s dawned on me that fft (pfft) shifting in max
> sounds like crap, excuse me for the vocals.
> there always seems to be a framesize drawback
> that i do not accept as proper sound.
> but altiverb for instance sounds quite accurate
> and doesnt have the mpeg like distortion to it.
>
> are there optimal settings for chunksize and overlap
> or.. am i just stuck with a poor sound with pfft

#112021
Sep 7, 2007 at 3:27pm

#112022
Sep 7, 2007 at 3:50pm

I always thought of them as beign called the ff”tease” like they tease you by sounding incredible and then crashing right when you get the perfect sound.

but they did get alot more stable in 2.5

#112023
Sep 7, 2007 at 7:14pm

lukas schrieb:
> it`s dawned on me that fft (pfft) shifting in max
> sounds like crap, excuse me for the vocals.

That means you want to pitch shift vocals? Then you have to deal with
formants as well and don’t shift them. This ain’t a trivial task and has
nothing to do with pfft~…
FFT is a mathematically pretty solid way of dealing with spectra. You
won’t be able to tweak the results if you don’t know how it works…
FFT in csound~, rtcmix, supercollider or Mathlab will sound exactly the
same if you do the same with it (same window, same size, same overlap,
same tweaks…)

Stefan


Stefan Tiedje————x——-
–_____———–|————–
–(_|_ —-|—–|—–()——-
– _|_)—-|—–()————–
———-()——–www.ccmix.com

#112024
Sep 7, 2007 at 11:44pm

Quote: Stefan Tiedje wrote on Fri, 07 September 2007 13:14
—————————————————-
> lukas schrieb:
> > it`s dawned on me that fft (pfft) shifting in max
> > sounds like crap, excuse me for the vocals.
>
> That means you want to pitch shift vocals? Then you have to deal with
> formants as well and don’t shift them. This ain’t a trivial task and has
> nothing to do with pfft~…

thats like altiverb, which also has nothing to do with fft.

#112025
Sep 8, 2007 at 12:16am

I think by vocals he meant his French (i.e. bad language!). Not that
he’s working with vocal sounds :-) Unless the comma was put in the
wrong place!

That aside the thing with csound and rtcmix is that the fft code is
already done for you so there is little DSP DIY involved. I have not
yet used rtcmix for fft but compared with max and supercollider
csound has given me the best result without going through much trouble.

And of course csound’s FFT is not great for realtime but then FFT is
very heavy and for a good result I wouldn’t use it realtime in max
either, but that’s just me… Maybe if I was a DSP guru it would be
different.

p

On 8 Sep 2007, at 00:44, Roman Thilenius wrote:

>
> Quote: Stefan Tiedje wrote on Fri, 07 September 2007 13:14
> —————————————————-
>> lukas schrieb:
>>> it`s dawned on me that fft (pfft) shifting in max
>>> sounds like crap, excuse me for the vocals.
>>
>> That means you want to pitch shift vocals? Then you have to deal with
>> formants as well and don’t shift them. This ain’t a trivial task
>> and has
>> nothing to do with pfft~…
>
>
> thats like altiverb, which also has nothing to do with fft.
> –
> http://vst-mac.info/

#112026
Sep 8, 2007 at 3:38pm

so i supose the oldschool fft windowed with a sine function instead of a buffer would sound the best?

#112027
Sep 8, 2007 at 4:41pm

lukas skrev:
> so i supose the oldschool fft windowed with a sine function instead of a buffer would sound the best?
I am sorry, but is that what you take away from the seven very good
responses you have been given?
Also, two of the replies also posed questions in order to help you
better. I think it would be prudent to answer those before we go any
further.

Just my two cents,
Andreas.

#112028
Sep 8, 2007 at 7:00pm

now.. lets suppose my da and ad work the same there must be a factor contributing to, the data sounding different.
i suppose its down to distortion, in the windowing department.

and thx for the externals:) always fun

#112029
Sep 8, 2007 at 7:22pm

> I always thought of them as being called the ff”tease” like they
> tease you by sounding incredible and then crashing right when
> you get the perfect sound.
>

We always thought of it as both fft “ease” and ff “tease” but “tease” like they tease you by sounding incredible and then you realize you just gobbled up 50% of the CPU capacity on your 400 MHz PowerBook ;)

> but they did get alot more stable in 2.5

One way to help me make them even more stable is to send crash reports when things go south.

Cheers,

Eric

#112030
Sep 8, 2007 at 8:13pm

#112031
Sep 8, 2007 at 8:39pm

peiman khosravi schrieb:
> That aside the thing with csound and rtcmix is that the fft code is
> already done for you so there is little DSP DIY involved.

This is also true for pfft~ in Max…

> I have not yet used rtcmix for fft but compared with max and
> supercollider csound has given me the best result without going
> through much trouble.

My trouble with csound is, that its way too low level for my laziness… ;-)

Stefan


Stefan Tiedje————x——-
–_____———–|————–
–(_|_ —-|—–|—–()——-
– _|_)—-|—–()————–
———-()——–www.ccmix.com

#112032
Sep 9, 2007 at 6:10pm

lukas schrieb:
> so i supose the oldschool fft windowed with a sine function instead
> of a buffer would sound the best?

I guess you mean a buffer filled with a windowing function?
Depends what you are after, what you want to do with the bins inside
pfft~… Using other windows is as oldschool as using hamming windows…

Stefan


Stefan Tiedje————x——-
–_____———–|————–
–(_|_ —-|—–|—–()——-
– _|_)—-|—–()————–
———-()——–www.ccmix.com

#112033
Sep 9, 2007 at 8:13pm

since when do fft~ and pfft~ use buffers with windowing functions in them?

#112034
Sep 9, 2007 at 8:15pm

#112035
Sep 9, 2007 at 8:18pm

#112036
Sep 9, 2007 at 11:21pm

Sorry am I missing something?. What exactly do you mean Altiverb has
“nothing” to do with FFT?…

On 08/09/2007, Roman Thilenius wrote:
>
> Quote: Stefan Tiedje wrote on Fri, 07 September 2007 13:14
> —————————————————-
> > lukas schrieb:
> > > it`s dawned on me that fft (pfft) shifting in max
> > > sounds like crap, excuse me for the vocals.
> >
> > That means you want to pitch shift vocals? Then you have to deal with
> > formants as well and don’t shift them. This ain’t a trivial task and has
> > nothing to do with pfft~…
>
>
> thats like altiverb, which also has nothing to do with fft.
> –
> http://vst-mac.info/
>


Pere Josep Villez
Creative and Computational Sound
Department of Creative Technologies
University of Portsmouth
36-40 Middle Street
Portsmouth
PO5
Tel 00 44 23 9284 8484

http://www.centuryofnoise.com
http://www.perevillez.com

#112037
Sep 10, 2007 at 2:31am

Quote: pvillez@gmail.com wrote on Sun, 09 September 2007 17:21
—————————————————-
> Sorry am I missing something?. What exactly do you mean Altiverb has
> “nothing” to do with FFT?…
>

okay, this was a really stupid comment of mine.

what i wanted to say is that one can hardly
compare the sound of a deconvolution process
with the sound of (another?) pitch shifting process,
because pitch shifting simply is not deconvolution.
you know, as in “apples and oranges”. :)

if you would multiply fft data in MSP in a deconvolution/reconvolution process it would
not create more artefacts than with the same
process using fft code from apple, lake, or fftw,
you just wold need a 50 GHz Mac to use it as reverb.

#112038
Sep 10, 2007 at 2:51am

OK but I will eat my computer if you can come up with a fft based
pitch scaling or time expansion in pure msp that sounds as good as,
say, audiosculpt, or even csound. I am sure it is possible but you
would have to do a year’s worth of coding probably.

Peiman

On 10 Sep 2007, at 03:31, Roman Thilenius wrote:

>
> Quote: pvillez@gmail.com wrote on Sun, 09 September 2007 17:21
> —————————————————-
>> Sorry am I missing something?. What exactly do you mean Altiverb has
>> “nothing” to do with FFT?…
>>
>
>
> okay, this was a really stupid comment of mine.
>
> what i wanted to say is that one can hardly
> compare the sound of a deconvolution process
> with the sound of (another?) pitch shifting process,
> because pitch shifting simply is not deconvolution.
> you know, as in “apples and oranges”. :)
>
> if you would multiply fft data in MSP in a deconvolution/
> reconvolution process it would
> not create more artefacts than with the same
> process using fft code from apple, lake, or fftw,
> you just wold need a 50 GHz Mac to use it as reverb.
>
>

#112039
Sep 10, 2007 at 3:42am

#112040
Sep 10, 2007 at 4:29am

> OK but I will eat my computer if you can come up with a fft based
> pitch scaling or time expansion in pure msp that sounds as good as,
> say, audiosculpt, or even csound. I am sure it is possible but you
> would have to do a year’s worth of coding probably.
>
> Peiman

audiosculpt has also not been made in 3 days. :)

as you might know, it does a bit more than changing the time in the fft bins, audioscupt is not a simple phase vocoder.

audiosculpt or spear as a creative tool play more in the resynthesis class than they would be linear processes.

the fourier transform itself is always the same, you must think of it as a math process like *~ and +~.

of course *~ and +~ will also sound better with 100 bits of precision at 192 Hz, as they will sound better when you interpolate modulation to avoid clicks, which explains why nuendo mixes sound better than mixes made in MSP.

#112041
Sep 10, 2007 at 9:38am

Pere Josep Villez schrieb:
> Sorry am I missing something?. What exactly do you mean…

Whenever I see this phrase, I check, and yes, its a reply to Roman… ;-)
You missed Roman, it has nothing to do with windows… or xease or…

Stefan


Stefan Tiedje————x——-
–_____———–|————–
–(_|_ —-|—–|—–()——-
– _|_)—-|—–()————–
———-()——–www.ccmix.com

#112042
Sep 10, 2007 at 1:01pm

i did manage to get a different sound from the oldschool:)
but that was due to a low frequency distortion created by the
sine window, so i suppose hanning cleared that issue.
not that dirty is bad for that matter.

only the externals do not pop up??.. what file do i need to put where, when the externals are put in the externals folder?

#112043
Sep 10, 2007 at 1:17pm

Yes I agree. But that is my point :-) To make an amazing sounding fft
based instrument with msp one needs to know all the math and DSP
stuff. It’s way more tricky than say making a granular instrument.
Now I am not a DSP guru and my knowledge of math is very limited so I
would rather just use an external such as csound~ or rtcmix~ and
spend more time composing than coding something that will never work
that well.

Best
Peiman

On 10 Sep 2007, at 05:29, Roman Thilenius wrote:

>
>
>> OK but I will eat my computer if you can come up with a fft based
>> pitch scaling or time expansion in pure msp that sounds as good as,
>> say, audiosculpt, or even csound. I am sure it is possible but you
>> would have to do a year’s worth of coding probably.
>>
>> Peiman
>
>
> audiosculpt has also not been made in 3 days. :)
>
> as you might know, it does a bit more than changing the time in the
> fft bins, audioscupt is not a simple phase vocoder.
>
> audiosculpt or spear as a creative tool play more in the
> resynthesis class than they would be linear processes.
>
> the fourier transform itself is always the same, you must think of
> it as a math process like *~ and +~.
>
> of course *~ and +~ will also sound better with 100 bits of
> precision at 192 Hz, as they will sound better when you interpolate
> modulation to avoid clicks, which explains why nuendo mixes sound
> better than mixes made in MSP.
>
>
>

#112044
Sep 10, 2007 at 3:15pm

Quote: Eric Lyon wrote on Sat, 08 September 2007 13:22
—————————————————-
> > I always thought of them as being called the ff”tease” like they
> > tease you by sounding incredible and then crashing right when
> > you get the perfect sound.
> >
>
> We always thought of it as both fft “ease” and ff “tease” but “tease” like they tease you by sounding incredible and then you realize you just gobbled up 50% of the CPU capacity on your 400 MHz PowerBook ;)
>
>
> > but they did get alot more stable in 2.5
>
> One way to help me make them even more stable is to send crash reports when things go south.
>
> Cheers,
>
> Eric
>
—————————————————-

Hi Eric,
I tried contacting you at the two e-mail addresses I found by Googling Eric Lyon FFTease and both bounced.
The e-mail here and the URL:
http://www.sarc.qub.ac.uk/~elyon/LyonSoftware/MaxMSP/FFTease/

http://arcana.dartmouth.edu/~eric/MAX/FFTease/

#112045
Sep 10, 2007 at 4:25pm

> Whenever I see this phrase, I check, and yes, its a reply to Roman… ;-)
> You missed Roman, it has nothing to do with windows… or xease or…
>

the truth is alwys hard to understand. :)

it is easily possible to build impulse response reverbs without any fourier transformation. hybrid models containing FIR only (plus other components such as allpasses) are quite hip a the moment.

because oyu have to pay for the petends for “zero latency” fft thats the way to go for many.

#112046
Sep 11, 2007 at 12:37pm

there shure are tricks to do a little enhancing.
but would`nt an fft based convolution verb sound
better then a fir.(if you are prepared throw the “realtime” aspects to the wind.

#112047
Nov 6, 2007 at 2:07pm

#112048

You must be logged in to reply to this topic.