DEF and PRF support?

You name it!
User avatar
dahnielson
Moderator
Posts: 632
Joined: Wed Jan 23, 2008 11:25 pm
Location: Linköping / Tranås, Sweden
Contact:

Re: DEF and PRF support?

Post by dahnielson » Tue Sep 09, 2008 9:24 pm

Andreas wrote:Anyway, I did a couple of measurements of the DEF filter a while ago, and ran it through prony. The result looked it was something like a 100 tap FIR filter, which made me give up.
Well, there must be some method to the madness. There must be a pattern in the frequency responses between filter settings. I guess we have to find another method to create a filter specification than prony.
Anders Dahnielson

Ardour2, Qtractor, Linuxsampler, M-AUDIO Delta 1010, Axiom 61, Korg D12, AKAI S2000, E-MU Proteus 2k, Roland R-5, Roland HP 1300e, Zoom RFX-1000, 4GB RAM x86_64 Intel Pentium Dual 1.80GHz Gentoo Linux

User avatar
dahnielson
Moderator
Posts: 632
Joined: Wed Jan 23, 2008 11:25 pm
Location: Linköping / Tranås, Sweden
Contact:

Re: DEF and PRF support?

Post by dahnielson » Wed Sep 10, 2008 11:58 am

I will see if I can prepare a sample/instrument with a sweep, record and deconvolve it to get the impulse response using Aliki. The IR should hopefully be of highest possible quality that way. I will post a GIG with it if that turn out to be the right tool. I'm curious to see what the DEF filter IR looks like in general.
Anders Dahnielson

Ardour2, Qtractor, Linuxsampler, M-AUDIO Delta 1010, Axiom 61, Korg D12, AKAI S2000, E-MU Proteus 2k, Roland R-5, Roland HP 1300e, Zoom RFX-1000, 4GB RAM x86_64 Intel Pentium Dual 1.80GHz Gentoo Linux

Andreas
Developer
Posts: 214
Joined: Sun Feb 03, 2008 8:33 am

Re: DEF and PRF support?

Post by Andreas » Wed Sep 10, 2008 5:33 pm

I found the files from the quick DEF tests I did. It seems as I tried three different settings: 7f-28-58-40, the supposedly flat one (doesn't look flat here), 7f-11-58-40 and 7f-28-58-36:

Image

User avatar
dahnielson
Moderator
Posts: 632
Joined: Wed Jan 23, 2008 11:25 pm
Location: Linköping / Tranås, Sweden
Contact:

Re: DEF and PRF support?

Post by dahnielson » Wed Sep 10, 2008 6:15 pm

Andreas wrote:I found the files from the quick DEF tests I did. It seems as I tried three different settings: 7f-28-58-40, the supposedly flat one (doesn't look flat here), 7f-11-58-40 and 7f-28-58-36
Whoa. Ok.

So back to decimal values and what they represent...

7f-28-58-40: Fc = 127, Q0 = 40, Q0H = 88, V0 = 64
7f-11-58-40: Fc = 127, Q0 = 17, Q0H = 88, V0 = 64
7f-28-58-36: Fc = 127, Q0 = 40, Q0H = 88, V0 = 54

The third setting shows clearly that V0 is independent and controls the overall gain (it's a "volume" control).

The difference between the first two settings are the Q0 value that "simultaneously determines gain, Quality (filter width) and a center frequency offset relative to the initial Fc setting". The plot show that for the maximum Fc setting (cut off frequency) lowering the Q0 value attenuate the frequencies around 10000 Hz distinctively more than the lower ones.

To me it looks like a regular sinc function (well, one half of it). But it's not a sinc filter as it is a frequency response plot in the frequency domain. (It would have been a low pass filter if it was a plot of the impulse response.)

If you lower the Fc0 setting I'll bet it start to look something like this:

Image

N.B. Q0 depend on Fc so something else might start to happen when Fc is lowered.
Anders Dahnielson

Ardour2, Qtractor, Linuxsampler, M-AUDIO Delta 1010, Axiom 61, Korg D12, AKAI S2000, E-MU Proteus 2k, Roland R-5, Roland HP 1300e, Zoom RFX-1000, 4GB RAM x86_64 Intel Pentium Dual 1.80GHz Gentoo Linux

User avatar
dahnielson
Moderator
Posts: 632
Joined: Wed Jan 23, 2008 11:25 pm
Location: Linköping / Tranås, Sweden
Contact:

Re: DEF and PRF support?

Post by dahnielson » Wed Sep 10, 2008 9:36 pm

Keeping in mind that both DEF/PRF probably are implemented by the same filter, the documentation tells us that it is supposed to be a formant filter. After taking a look at the Format Filtering Example in Introduction to Digital Filters with Audio Applications, especially the plot at the bottom of the page, I can wage a guess that's similar to what is going on in your frequency response plot:

Image
Anders Dahnielson

Ardour2, Qtractor, Linuxsampler, M-AUDIO Delta 1010, Axiom 61, Korg D12, AKAI S2000, E-MU Proteus 2k, Roland R-5, Roland HP 1300e, Zoom RFX-1000, 4GB RAM x86_64 Intel Pentium Dual 1.80GHz Gentoo Linux

User avatar
dahnielson
Moderator
Posts: 632
Joined: Wed Jan 23, 2008 11:25 pm
Location: Linköping / Tranås, Sweden
Contact:

Re: DEF and PRF support?

Post by dahnielson » Thu Sep 11, 2008 10:43 am

dahnielson wrote:Keeping in mind that both DEF/PRF probably are implemented by the same filter, the documentation tells us that it is supposed to be a formant filter.
D'oh! Of course it is the same filter. It's a 7th order formant filter, meaning seven setting to interpolate between: four PRF and three DEF, so that you can both transform a mf sample to f and do a glissando with the same filter at the same time. I'm slow sometimes.
dahnielson wrote:After taking a look at the Format Filtering Example in Introduction to Digital Filters with Audio Applications, especially the plot at the bottom of the page, I can wage a guess that's similar to what is going on in your frequency response plot
The fact that it's a formant filter makes the description of Fc and Q obvious, now when I've read up on the subject. Fc sets the center frequency for F1 while Q control the "center frequency offset relative to the initial Fc setting" for F2. Q probably control the bandwidth for both F1 and F2. Q0H control the bandwidth of F3 with a fixed center frequency at 7.5 kHz. There is no need to specify gains directly:

"[The] transfer function is an all-pole filter [...]. As a result, there is no need to specify gains for the formant resonators -- only center-frequency and bandwidth are necessary to specify each formant, leaving only an overall scale factor unspecified in a cascade (series) formant filter bank."

The formant filters can be implemented in parallel rather than series:

"In principle, the formant filter sections are in series, [...] [n]umerically, however, it makes more sense to implement disjoint resonances in parallel rather than in series. This is because when one formant filter is resonating, the others will be attenuating, so that to achieve a particular peak-gain at resonance, the resonating filter must overcome all combined attenuations as well as applying its own gain."

(But I still think your plot looks a lot like a sinc function.)

It would be great if you could do additional measurements to find out if my assumptions are correct.
Anders Dahnielson

Ardour2, Qtractor, Linuxsampler, M-AUDIO Delta 1010, Axiom 61, Korg D12, AKAI S2000, E-MU Proteus 2k, Roland R-5, Roland HP 1300e, Zoom RFX-1000, 4GB RAM x86_64 Intel Pentium Dual 1.80GHz Gentoo Linux

Post Reply