[music-dsp] prime number's role in reverb

[y***@gpe-hkg.com]

Dear list:
I have improve the ring noise by modulating the delay line. Sorry to bore
you.
Yet i have another question about prime number:
Most article mentioned that the delay length of FDN reverb must be prime
number,for example,919,1129,etc.
Today,i expermented on three sets of delay line,as below:
The first delay sets is pure prime number:
m=[919 997 1061 1093 1129 1151 1171 1187 1213 1237 1259 1283 1303 1319 1327
1361]
The second sets is not prime number,but similar to set 1:
m=[900 990 1060 1090 1120 1150 1170 1180 1210 1230 1250 1280 1300 1310
1320 1360]
The third sets is definitly regular even number:
m=[900 950 1000 1050 1100 1150 1200 1250 1300 1350 1400 1450 1500 1550
1600 1650]
I implement them in Csound, each delay line is modulated by LFO slightly
(0.1Hz).The dry source included drum and a piece of classic guitar solo.
All processed result are saved to separated wave files.The monitor device
is a Mona Echo 2496 soundcard and Event 20/20 loudspeaker.
To my surprise,I indeed could not find any difference from the three
sets,as well as the spectrum and decay feature (observed in Cooledit Pro's
Frequency Analysis window).Each wave sounds just the same to me!
But, why? The prime number have what role in reverb?

Thanks and Best Begards!

+---------------------------------------------+
| Shi-Yong (师勇) |
| R&D,GPE Limited,Shenzhen,China |
| Tel: [phone] ext. 380 |
| Fax: [phone] |
| mail to: ***@gpe-hkg.com |
+---------------------------------------------+





***@gpe-hkg.com
Sent by: To: music-***@aulos.calarts.edu
music-dsp-***@aulos.c cc:
alarts.edu Subject: [music-dsp] "ring" tail in FDN reverb


03-07-08 17:20
Please respond to
music-dsp







Dear list:
I'm tring a 32 delay lines FDN algorithm in Csound just now.The
processing speed is very slow on my P4,and the sound quality is not much
better than 16 delay lines.A metallic "ring" noise about 2.5kHz can be
heard in the reverb tail in some audio sample (guitar solo).
I don't know why.The delay line I used is below:
m=[1693 1723 1747 1777 1801 1823 1847 1873 1901 1931 1951 1979 2003 2029
2053 2081 2111 2137 2153 2179 2203 2237 2267 2293 2311 2339 2357 2381 2411
2437 2459 2503]
Should I try more different sets of delay line length again and
again? Any suggestion?
Thanks!

Best Regards!

+---------------------------------------------+
| Shi-Yong (师勇) |
| R&D,GPE Limited,Shenzhen,China |
| Tel: [phone] ext. 380 |
| Fax: [phone] |
| mail to: ***@gpe-hkg.com |
+---------------------------------------------+



dupswapdrop -- the music-dsp mailing list and website: subscription info,
FAQ, source code archive, list archive, book reviews, dsp links
http://shoko.calarts.edu/musicdsp/
http://ceait.calarts.edu/mailman/listinfo/music-dsp

[joshua reich]

When you say your are modulating each delay line, are you modulating the
delay length, or its gain as it feeds into the fdn ?

If you are modulating the delay length then the similarity described seems
to make sense to me.

On Thu, 10 Jul 2003 ***@gpe-hkg.com wrote:

[snip]

Post by y***@gpe-hkg.com
I implement them in Csound, each delay line is modulated by LFO slightly
(0.1Hz).The dry source included drum and a piece of classic guitar solo.
All processed result are saved to separated wave files.The monitor device
is a Mona Echo 2496 soundcard and Event 20/20 loudspeaker.
To my surprise,I indeed could not find any difference from the three
sets,as well as the spectrum and decay feature (observed in Cooledit Pro's
Frequency Analysis window).Each wave sounds just the same to me!
But, why? The prime number have what role in reverb?
Thanks and Best Begards!
+---------------------------------------------+
| Shi-Yong (ʦ��) |
| R&D,GPE Limited,Shenzhen,China |
| Tel: [phone] ext. 380 |
| Fax: [phone] |
+---------------------------------------------+
alarts.edu Subject: [music-dsp] "ring" tail in FDN reverb
03-07-08 17:20
Please respond to
music-dsp
I'm tring a 32 delay lines FDN algorithm in Csound just now.The
processing speed is very slow on my P4,and the sound quality is not much
better than 16 delay lines.A metallic "ring" noise about 2.5kHz can be
heard in the reverb tail in some audio sample (guitar solo).
m=[1693 1723 1747 1777 1801 1823 1847 1873 1901 1931 1951 1979 2003 2029
2053 2081 2111 2137 2153 2179 2203 2237 2267 2293 2311 2339 2357 2381 2411
2437 2459 2503]
Should I try more different sets of delay line length again and
again? Any suggestion?
Thanks!
Best Regards!
+---------------------------------------------+
| Shi-Yong (ʦ��) |
| R&D,GPE Limited,Shenzhen,China |
| Tel: [phone] ext. 380 |
| Fax: [phone] |
+---------------------------------------------+
dupswapdrop -- the music-dsp mailing list and website: subscription info,
FAQ, source code archive, list archive, book reviews, dsp links
http://shoko.calarts.edu/musicdsp/
http://ceait.calarts.edu/mailman/listinfo/music-dsp
dupswapdrop -- the music-dsp mailing list and website: subscription info, FAQ, source code archive, list archive, book reviews, dsp links http://shoko.calarts.edu/musicdsp/
http://ceait.calarts.edu/mailman/listinfo/music-dsp

[Svante Stadler]

They probably use prime numbers to minimize the risk that a subset of the
delay lines are harmonically related, thus giving a metallic ring to the
reverb (just guessing here). That doesn't make much sense though, as the
resonance frequencies are proportional to the inverse of the delay lengths.

/Svante

Post by y***@gpe-hkg.com
I have improve the ring noise by modulating the delay line. Sorry to bore
you.
Most article mentioned that the delay length of FDN reverb must be prime
number,for example,919,1129,etc.
m=[919 997 1061 1093 1129 1151 1171 1187 1213 1237 1259 1283 1303 1319 1327
1361]
m=[900 990 1060 1090 1120 1150 1170 1180 1210 1230 1250 1280 1300 1310
1320 1360]
m=[900 950 1000 1050 1100 1150 1200 1250 1300 1350 1400 1450 1500 1550
1600 1650]
I implement them in Csound, each delay line is modulated by LFO slightly
(0.1Hz).The dry source included drum and a piece of classic guitar solo.
All processed result are saved to separated wave files.The monitor device
is a Mona Echo 2496 soundcard and Event 20/20 loudspeaker.
To my surprise,I indeed could not find any difference from the three
sets,as well as the spectrum and decay feature (observed in Cooledit Pro's
Frequency Analysis window).Each wave sounds just the same to me!
But, why? The prime number have what role in reverb?
Thanks and Best Begards!
+---------------------------------------------+
| Shi-Yong (ÊŠÓÂ) |
| R&D,GPE Limited,Shenzhen,China |
| Tel: [phone] ext. 380 |
| Fax: [phone] |
+---------------------------------------------+
alarts.edu Subject: [music-dsp]
"ring" tail in FDN reverb
03-07-08 17:20
Please respond to
music-dsp
I'm tring a 32 delay lines FDN algorithm in Csound just now.The
processing speed is very slow on my P4,and the sound quality is not much
better than 16 delay lines.A metallic "ring" noise about 2.5kHz can be
heard in the reverb tail in some audio sample (guitar solo).
m=[1693 1723 1747 1777 1801 1823 1847 1873 1901 1931 1951 1979 2003 2029
2053 2081 2111 2137 2153 2179 2203 2237 2267 2293 2311 2339 2357 2381 2411
2437 2459 2503]
Should I try more different sets of delay line length again and
again? Any suggestion?
Thanks!
Best Regards!
+---------------------------------------------+
| Shi-Yong (ÊŠÓÂ) |
| R&D,GPE Limited,Shenzhen,China |
| Tel: [phone] ext. 380 |
| Fax: [phone] |
+---------------------------------------------+
dupswapdrop -- the music-dsp mailing list and website: subscription info,
FAQ, source code archive, list archive, book reviews, dsp links
http://shoko.calarts.edu/musicdsp/
http://ceait.calarts.edu/mailman/listinfo/music-dsp
dupswapdrop -- the music-dsp mailing list and website: subscription info,
FAQ, source code archive, list archive, book reviews, dsp links
http://shoko.calarts.edu/musicdsp/
http://ceait.calarts.edu/mailman/listinfo/music-dsp

_________________________________________________________________
The new MSN 8: smart spam protection and 2 months FREE*
http://join.msn.com/?page=features/junkmail

[James McCartney]

Yes. Prime numbers can still be in arbitrarily close approximations to
harmonic (small integer) ratios (e.g. 2999/997 = 3.008). So just
choosing some prime numbers is not sufficient. Also really not
necessary if you are careful.

Post by Svante Stadler
They probably use prime numbers to minimize the risk that a subset of
the delay lines are harmonically related

[James McCartney]

Yes. Prime numbers can still be in arbitrarily close approximations to
harmonic (small integer) ratios (e.g. 2999/997 = 3.008). So just
choosing some prime numbers is not sufficient. Also really not
necessary if you are careful.

Post by Svante Stadler
They probably use prime numbers to minimize the risk that a subset of
the delay lines are harmonically related

[Jens Blomquist]

Well, I'm no expert but I've implemented a couple of reverbs.

May I ask you a couple of questions?

1. You name the modulation speed, but what about the amount?
2. Same modulation for all delay lines?
3. How does the feedback matrix look like?

The choise for prime numbers should be pretty obvious in theory.
Look at the third set: 1000*3 = 1500*2.

In practice? I don't know. The best sounding reverbs I have
created is far from realistic. I usually go crazy with the
modulation as I like the "unrealistic" thick density better
than the dull and often metallic sounding reverbs out there.

Also, things get rather complicated when using modulation. Maby
someone here can tell us what difference prime numbers will
make in such a case?

br
Jens



----- Original Message -----
From: <***@gpe-hkg.com>
To: <music-***@aulos.calarts.edu>
Sent: Thursday, July 10, 2003 11:39 AM
Subject: [music-dsp] prime number's role in reverb

Post by y***@gpe-hkg.com
I have improve the ring noise by modulating the delay line. Sorry to bore
you.
Most article mentioned that the delay length of FDN reverb must be prime
number,for example,919,1129,etc.
m=[919 997 1061 1093 1129 1151 1171 1187 1213 1237 1259 1283 1303 1319 1327
1361]
m=[900 990 1060 1090 1120 1150 1170 1180 1210 1230 1250 1280 1300 1310
1320 1360]
m=[900 950 1000 1050 1100 1150 1200 1250 1300 1350 1400 1450 1500 1550
1600 1650]
I implement them in Csound, each delay line is modulated by LFO slightly
(0.1Hz).The dry source included drum and a piece of classic guitar solo.
All processed result are saved to separated wave files.The monitor device
is a Mona Echo 2496 soundcard and Event 20/20 loudspeaker.
To my surprise,I indeed could not find any difference from the three
sets,as well as the spectrum and decay feature (observed in Cooledit Pro's
Frequency Analysis window).Each wave sounds just the same to me!
But, why? The prime number have what role in reverb?
Thanks and Best Begards!
+---------------------------------------------+
| Shi-Yong (师勇) |
| R&D,GPE Limited,Shenzhen,China |
| Tel: [phone] ext. 380 |
| Fax: [phone] |
+---------------------------------------------+
alarts.edu Subject: [music-dsp]

"ring" tail in FDN reverb

Post by y***@gpe-hkg.com
03-07-08 17:20
Please respond to
music-dsp
I'm tring a 32 delay lines FDN algorithm in Csound just now.The
processing speed is very slow on my P4,and the sound quality is not much
better than 16 delay lines.A metallic "ring" noise about 2.5kHz can be
heard in the reverb tail in some audio sample (guitar solo).
m=[1693 1723 1747 1777 1801 1823 1847 1873 1901 1931 1951 1979 2003 2029
2053 2081 2111 2137 2153 2179 2203 2237 2267 2293 2311 2339 2357 2381 2411
2437 2459 2503]
Should I try more different sets of delay line length again and
again? Any suggestion?
Thanks!
Best Regards!
+---------------------------------------------+
| Shi-Yong (师勇) |
| R&D,GPE Limited,Shenzhen,China |
| Tel: [phone] ext. 380 |
| Fax: [phone] |
+---------------------------------------------+
dupswapdrop -- the music-dsp mailing list and website: subscription info,
FAQ, source code archive, list archive, book reviews, dsp links
http://shoko.calarts.edu/musicdsp/
http://ceait.calarts.edu/mailman/listinfo/music-dsp
dupswapdrop -- the music-dsp mailing list and website: subscription info,

FAQ, source code archive, list archive, book reviews, dsp links
http://shoko.calarts.edu/musicdsp/

Post by y***@gpe-hkg.com
http://ceait.calarts.edu/mailman/listinfo/music-dsp

[y***@gpe-hkg.com]

Yes,i am modulating the delay length. Although i also have tried to
modulate the feedback gain,but sounds not very comfortable,so i stop.

Best Regards!

+---------------------------------------------+
| Shi-Yong (师勇) |
| R&D,GPE Limited,Shenzhen,China |
| Tel: [phone] ext. 380 |
| Fax: [phone] |
| mail to: ***@gpe-hkg.com |
+---------------------------------------------+





joshua reich
<***@i2pi.com> To: <music-***@aulos.calarts.edu>
Sent by: cc:
music-dsp-***@aulos.c Subject: Re: [music-dsp] prime number's role in reverb
alarts.edu


03-07-10 19:36
Please respond to
music-dsp








When you say your are modulating each delay line, are you modulating the
delay length, or its gain as it feeds into the fdn ?

If you are modulating the delay length then the similarity described seems
to make sense to me.

On Thu, 10 Jul 2003 ***@gpe-hkg.com wrote:

[snip]

Post by y***@gpe-hkg.com
I implement them in Csound, each delay line is modulated by LFO slightly
(0.1Hz).The dry source included drum and a piece of classic guitar solo.
All processed result are saved to separated wave files.The monitor device
is a Mona Echo 2496 soundcard and Event 20/20 loudspeaker.
To my surprise,I indeed could not find any difference from the three
sets,as well as the spectrum and decay feature (observed in Cooledit

Pro's

Post by y***@gpe-hkg.com
Frequency Analysis window).Each wave sounds just the same to me!
But, why? The prime number have what role in reverb?
Thanks and Best Begards!
+---------------------------------------------+
| Shi-Yong (师勇) |
| R&D,GPE Limited,Shenzhen,China |
| Tel: [phone] ext. 380 |
| Fax: [phone] |
+---------------------------------------------+
alarts.edu Subject: [music-dsp]

"ring" tail in FDN reverb

Post by y***@gpe-hkg.com
03-07-08 17:20
Please respond to
music-dsp
I'm tring a 32 delay lines FDN algorithm in Csound just now.The
processing speed is very slow on my P4,and the sound quality is not much
better than 16 delay lines.A metallic "ring" noise about 2.5kHz can be
heard in the reverb tail in some audio sample (guitar solo).
m=[1693 1723 1747 1777 1801 1823 1847 1873 1901 1931 1951 1979 2003 2029
2053 2081 2111 2137 2153 2179 2203 2237 2267 2293 2311 2339 2357 2381 2411
2437 2459 2503]
Should I try more different sets of delay line length again and
again? Any suggestion?
Thanks!
Best Regards!
+---------------------------------------------+
| Shi-Yong (师勇) |
| R&D,GPE Limited,Shenzhen,China |
| Tel: [phone] ext. 380 |
| Fax: [phone] |
+---------------------------------------------+
dupswapdrop -- the music-dsp mailing list and website: subscription info,
FAQ, source code archive, list archive, book reviews, dsp links
http://shoko.calarts.edu/musicdsp/
http://ceait.calarts.edu/mailman/listinfo/music-dsp
dupswapdrop -- the music-dsp mailing list and website: subscription info,

FAQ, source code archive, list archive, book reviews, dsp links
http://shoko.calarts.edu/musicdsp/

Post by y***@gpe-hkg.com
http://ceait.calarts.edu/mailman/listinfo/music-dsp

--
joshua reich

***@i2pi.com
Ph: [phone]
Mb: [phone]



dupswapdrop -- the music-dsp mailing list and website: subscription info,
FAQ, source code archive, list archive, book reviews, dsp links
http://shoko.calarts.edu/musicdsp/
http://ceait.calarts.edu/mailman/listinfo/music-dsp

[y***@gpe-hkg.com]

Hi,Jens,
Part of the Csound code is like this:(different modulation for each delay
line).
If let the iLFO_Dph bigger than 0.1,the guitar reverb will be terrible.

iLFO_Dph = 0.01 ;modulation amount
iLFO_Rate = 0.1 ;modulation rate
im1 = 919
im2 = 997
im3 = 1061
...
idelt1 = im1/sr
idelt2 = im2/sr
idelt3 = im3/sr
...

at1 oscil idelt1*iLFO_Dph, iLFO_Rate*0.50, 2, .2
at2 oscil idelt2*iLFO_Dph, iLFO_Rate*0.56, 2, .4
at3 oscil idelt3*iLFO_Dph, iLFO_Rate*0.54, 2, .6
...
at16 oscil idelt16*iLFO_Dph, iLFO_Rate*0.59, 2, .55

adelt1 = (idelt1+at1)*1000
adelt2 = (idelt2+at2)*1000
adelt3 = (idelt3+at3)*1000
...
adelt16 = (idelt16+at16)*1000

Best Regards!

+---------------------------------------------+
| Shi-Yong (师勇) |
| R&D,GPE Limited,Shenzhen,China |
| Tel: [phone] ext. 380 |
| Fax: [phone] |
| mail to: ***@gpe-hkg.com |
+---------------------------------------------+





"Jens Blomquist"
<***@telia.com> To: <music-***@aulos.calarts.edu>
Sent by: cc:
music-dsp-***@aulos.c Subject: Re: [music-dsp] prime number's role in reverb
alarts.edu


03-07-11 07:41
Please respond to
music-dsp







Well, I'm no expert but I've implemented a couple of reverbs.

May I ask you a couple of questions?

1. You name the modulation speed, but what about the amount?
2. Same modulation for all delay lines?
3. How does the feedback matrix look like?

The choise for prime numbers should be pretty obvious in theory.
Look at the third set: 1000*3 = 1500*2.

In practice? I don't know. The best sounding reverbs I have
created is far from realistic. I usually go crazy with the
modulation as I like the "unrealistic" thick density better
than the dull and often metallic sounding reverbs out there.

Also, things get rather complicated when using modulation. Maby
someone here can tell us what difference prime numbers will
make in such a case?

br
Jens



----- Original Message -----
From: <***@gpe-hkg.com>
To: <music-***@aulos.calarts.edu>
Sent: Thursday, July 10, 2003 11:39 AM
Subject: [music-dsp] prime number's role in reverb

Post by y***@gpe-hkg.com
I have improve the ring noise by modulating the delay line. Sorry to bore
you.
Most article mentioned that the delay length of FDN reverb must be prime
number,for example,919,1129,etc.
m=[919 997 1061 1093 1129 1151 1171 1187 1213 1237 1259 1283 1303 1319

1327

Post by y***@gpe-hkg.com
1361]
m=[900 990 1060 1090 1120 1150 1170 1180 1210 1230 1250 1280 1300 1310
1320 1360]
m=[900 950 1000 1050 1100 1150 1200 1250 1300 1350 1400 1450 1500 1550
1600 1650]
I implement them in Csound, each delay line is modulated by LFO slightly
(0.1Hz).The dry source included drum and a piece of classic guitar solo.
All processed result are saved to separated wave files.The monitor device
is a Mona Echo 2496 soundcard and Event 20/20 loudspeaker.
To my surprise,I indeed could not find any difference from the three
sets,as well as the spectrum and decay feature (observed in Cooledit

Pro's

Post by y***@gpe-hkg.com
Frequency Analysis window).Each wave sounds just the same to me!
But, why? The prime number have what role in reverb?
Thanks and Best Begards!
+---------------------------------------------+
| Shi-Yong (师勇) |
| R&D,GPE Limited,Shenzhen,China |
| Tel: [phone] ext. 380 |
| Fax: [phone] |
+---------------------------------------------+

[music-dsp]
"ring" tail in FDN reverb

Post by y***@gpe-hkg.com
03-07-08 17:20
Please respond to
music-dsp
I'm tring a 32 delay lines FDN algorithm in Csound just now.The
processing speed is very slow on my P4,and the sound quality is not much
better than 16 delay lines.A metallic "ring" noise about 2.5kHz can be
heard in the reverb tail in some audio sample (guitar solo).
m=[1693 1723 1747 1777 1801 1823 1847 1873 1901 1931 1951 1979 2003 2029
2053 2081 2111 2137 2153 2179 2203 2237 2267 2293 2311 2339 2357 2381

2411

Post by y***@gpe-hkg.com
2437 2459 2503]
Should I try more different sets of delay line length again and
again? Any suggestion?
Thanks!
Best Regards!
+---------------------------------------------+
| Shi-Yong (师勇) |
| R&D,GPE Limited,Shenzhen,China |
| Tel: [phone] ext. 380 |
| Fax: [phone] |
+---------------------------------------------+
dupswapdrop -- the music-dsp mailing list and website: subscription info,
FAQ, source code archive, list archive, book reviews, dsp links
http://shoko.calarts.edu/musicdsp/
http://ceait.calarts.edu/mailman/listinfo/music-dsp
dupswapdrop -- the music-dsp mailing list and website: subscription info,

FAQ, source code archive, list archive, book reviews, dsp links
http://shoko.calarts.edu/musicdsp/

Post by y***@gpe-hkg.com
http://ceait.calarts.edu/mailman/listinfo/music-dsp

dupswapdrop -- the music-dsp mailing list and website: subscription info,
FAQ, source code archive, list archive, book reviews, dsp links
http://shoko.calarts.edu/musicdsp/
http://ceait.calarts.edu/mailman/listinfo/music-dsp

[Juhana Sadeharju]

Post by y***@gpe-hkg.com
I have improve the ring noise by modulating the delay line. Sorry to bore
you.

Maybe you should first reduce the ringing to minimum, and only
after then add modulations for further improvements.

Post by y***@gpe-hkg.com
But, why? The prime number have what role in reverb?

Only reason I have figured out is that the distinct echoes do
not overlap too soon. "Too soon" is relevant as the echoes do
overlap eventually. The prime rule seems to be weaker than
my optimization rule: distinct echoes should not become too near
too soon. Because even if the echoes would be at prime numbers,
they may be very close together.

One could compute the delay line lengths, as Rocchesso did, from
the room normal modes. I had some approximate formula which
I posted here years ago. Check Csound babo reverb for values
and formulae.

Please keep us (or me) informed on how you success.

Best regards,

Juhana

[joshua reich]

Well, if you are modulating the delay length, then any advantages of using
prime numbers will be lost. As others have mentioned the basic rationale
behind using prime number delay lengths is to minimise the correlation
between delay length and factors in the period length of your incoming
signal (thus reducing the liklihood of ringing). If you are modulate the
length, then you are only using a prime number a small amount of the time.
For example if your delay length is 10007, modulated +/- 5, then only 2 of
the lengths in the range are prime, with all the others having a large
number of factors that may or may not cause ringing.

As James McCartney says, and my small experience playing with FDN's
confirms, you dont _need_ prime delay lengths. The problem of eliminating
ringing is one common to all reverb algorithms im aware of, and i have yet
to find any good sources explaining 'analytic' ways of minimisation. It
always seems to come down to trial and error.

Joshua Reich.

Post by y***@gpe-hkg.com
Yes,i am modulating the delay length. Although i also have tried to
modulate the feedback gain,but sounds not very comfortable,so i stop.
Best Regards!
+---------------------------------------------+
| Shi-Yong (ʦ��) |
| R&D,GPE Limited,Shenzhen,China |
| Tel: [phone] ext. 380 |
| Fax: [phone] |
+---------------------------------------------+

[Richard Dobson]

I would have thouht that these days, people would be using (some) fractional
delay lines (with more or less sophisticated interpolation); so that the issue
of integer delay lengths is replaceed by other delay length issues.
??


Richard Dobson

Post by joshua reich
Well, if you are modulating the delay length, then any advantages of using
prime numbers will be lost. As others have mentioned the basic rationale
behind using prime number delay lengths is to minimise the correlation
between delay length and factors in the period length of your incoming
signal (thus reducing the liklihood of ringing). If you are modulate the
length, then you are only using a prime number a small amount of the time.
For example if your delay length is 10007, modulated +/- 5, then only 2 of
the lengths in the range are prime, with all the others having a large
number of factors that may or may not cause ringing.
As James McCartney says, and my small experience playing with FDN's
confirms, you dont _need_ prime delay lengths. The problem of eliminating
ringing is one common to all reverb algorithms im aware of, and i have yet
to find any good sources explaining 'analytic' ways of minimisation. It
always seems to come down to trial and error.
Joshua Reich.

[Vesa Norilo]

Post by Richard Dobson
I would have thouht that these days, people would be using (some)
fractional delay lines (with more or less sophisticated interpolation);
so that the issue of integer delay lengths is replaceed by other delay
length issues.

Richard,

What are the advantages of fractional delays in reverberation? I can
only think of two disadvantages, cpu load and artifacts from imperfect
interpolation.

Vesa

[Richard Dobson]

I haven't tried it myself (yet), but there is the suggestion to modulate the
delay times to help minimize ringing effects, etc.

http://www-ccrma.stanford.edu/~jos/waveguide/Time_Varying_Reverberators.html


I don't think one would want to do that without at least linear interpolation,
unless the delays were really long already. I would have thought interpolation
artefacts would not cause a huge problem in this sort of application. The main
artefact of linear interpolation would, I would guess, be a slight HF loss,and
that would be covered by any filter in the feedback loop anyway. Maybe it was
suggested, but nobody actually does it that way?

</guess>

Richard Dobson

Post by Vesa Norilo

Post by Richard Dobson
I would have thouht that these days, people would be using (some)
fractional delay lines (with more or less sophisticated
interpolation); so that the issue of integer delay lengths is
replaceed by other delay length issues.

Richard,
What are the advantages of fractional delays in reverberation? I can
only think of two disadvantages, cpu load and artifacts from imperfect
interpolation.
Vesa

[Alexis Glass]

I think there was some research done by people at the Helsinky
University of Technology (probably Valimaki and Karjalainen, but I'm not
sure) wherein they demonstrated that FIR (specifically Lagrange)
interpolators were most appropriate for waveguides and waveguide meshes
because they didn't cause the transients that all pass filter-based
interpolators tended to make. I don't see why the same wouldn't hold
true for FDN-based reverberation algorithms.

A useful paper is online here:
http://www.acoustics.hut.fi/~vpv/publications/icassp00-fd.pdf

Alexis Glass

Post by Richard Dobson
I haven't tried it myself (yet), but there is the suggestion to
modulate the delay times to help minimize ringing effects, etc.
http://www-ccrma.stanford.edu/~jos/waveguide/Time_Varying_Reverberators.html
I don't think one would want to do that without at least linear
interpolation, unless the delays were really long already. I would
have thought interpolation artefacts would not cause a huge problem in
this sort of application. The main artefact of linear interpolation
would, I would guess, be a slight HF loss,and that would be covered by
any filter in the feedback loop anyway. Maybe it was suggested, but
nobody actually does it that way?
</guess>
Richard Dobson

[joshua reich]

Admittedly i havent put much time into optimisation of fdn code, but i
found that moving to fractional delays became too processor intensive,
especially with higher order interpolation - and as you say, you just run
into a different set of issues.

Post by Richard Dobson
I would have thouht that these days, people would be using (some) fractional
delay lines (with more or less sophisticated interpolation); so that the issue
of integer delay lengths is replaceed by other delay length issues.
??
Richard Dobson

[James Chandler Jr]

Post by Richard Dobson
I would have thouht that these days, people would be using (some)
fractional delay lines (with more or less sophisticated
interpolation); so that the issue of integer delay lengths is
replaceed by other delay length issues.
??

Hi Richard

I may have stupidly misinterpreted the old recommendations for
prime-related delay lines, but had assumed the advice referred to prime
ratios between delay times, rather than a prime number of samples per
delay line.

As was already mentioned, the larger primes suitable for sample delay
lengths, can accidentally be pretty close to integer delay ratios and
encourage a metallic thin sound.

A careful selection of small-prime delay ratios might give better
"spread" to the delays, helping minimize the tendency of echoes from
several taps accidentally reinforcing each other.

I tried prime ratios for delay taps, but primes didn't seem to have an
inherent advantage over other delay ratios designed to be "non
integral". Designing the taps and other reverb features is black art.
Small differences in algorithm and settings can spell the difference
between good and bad-sounding reverb.

Regardless of tap delay ratio, the echoes will reinforce at some point
during the decay tail. If a ratio is something like 1 : 1.1 -- the
echoes start out close-spaced, diverge, then re-converge again. Same
deal for any non-integral ratio.

One could design a non-integral set of multiple tap ratios that "ought
to" be fat, but in practice the reverb might start fat, quickly evolve
to thin, and then diverge to fat again, giving a pretty unsatisfactory
tail characteristic.

James Chandler Jr.

[Richard Dobson]

No, the rule about mutually prime delays is the one given, so you are right
there. ~My~ assumption is that this must be ensured by making each delay a prime
number of samples (I even have some code that receives a delay in seconds, and
converts it into the nearest prime); but I have also read that choosing a prime
number 2 samples longer, for example, can make a great difference. Possibly that
relates to the almost-integer ratio problem. Everyone says reverb creation is
still at least as much art as science! Still hope for me then...:-)


Richard Dobson

Post by James Chandler Jr

Post by Richard Dobson
I would have thouht that these days, people would be using (some)
fractional delay lines (with more or less sophisticated
interpolation); so that the issue of integer delay lengths is
replaceed by other delay length issues.
??

Hi Richard
I may have stupidly misinterpreted the old recommendations for
prime-related delay lines, but had assumed the advice referred to prime
ratios between delay times, rather than a prime number of samples per
delay line.
As was already mentioned, the larger primes suitable for sample delay
lengths, can accidentally be pretty close to integer delay ratios and
encourage a metallic thin sound.

....

[Frederick Umminger]

----- Original Message -----

Post by James Chandler Jr
Regardless of tap delay ratio, the echoes will reinforce at some point
during the decay tail. If a ratio is something like 1 : 1.1 -- the
echoes start out close-spaced, diverge, then re-converge again. Same
deal for any non-integral ratio.
One could design a non-integral set of multiple tap ratios that "ought
to" be fat, but in practice the reverb might start fat, quickly evolve
to thin, and then diverge to fat again, giving a pretty unsatisfactory
tail characteristic.
James Chandler Jr.

If two comb filters c1 and c2 have delays d1 and d2, then they phase from
fat to thin and back at the beat frequency 1/d1-1/d2. Similarly comb filter
c3 will phase with c2 at the beat frequency 1/d3 - 1/d2. By choosing
1/d3-1/d2 = 1/d2-1/d1 (i.e. d3 = d1 d2 / (2*d1-d2)), these beat frequencies
can be made equal. Then, in theory, an appropriate predelay before c3 can be
used to set the beats 180 degrees out of phase to make the density remain
roughly constant.
This idea can be extended to any number of combs. In practice, however, it
doesn't seem to really work all that well.

-Frederick Umminger


________________________________________________________________________
Delivered using the Free Personal Edition of Mailtraq (www.mailtraq.com)

[James Chandler Jr]

Post by Jens Blomquist
----- Original Message -----

Post by James Chandler Jr
One could design a non-integral set of multiple tap ratios that "ought
to" be fat, but in practice the reverb might start fat, quickly evolve
to thin, and then diverge to fat again, giving a pretty unsatisfactory
tail characteristic.

If two comb filters c1 and c2 have delays d1 and d2, then they phase
from
fat to thin and back at the beat frequency 1/d1-1/d2. Similarly comb
filter
c3 will phase with c2 at the beat frequency 1/d3 - 1/d2. By choosing
1/d3-1/d2 = 1/d2-1/d1 (i.e. d3 = d1 d2 / (2*d1-d2)), these beat
frequencies
can be made equal. Then, in theory, an appropriate predelay before c3
can be
used to set the beats 180 degrees out of phase to make the density
remain
roughly constant.
This idea can be extended to any number of combs. In practice,
however, it
doesn't seem to really work all that well.

Hi Frederick

Thanks. Sounds promising, for algorithmically picking comb delay times
in order to "smooth out" a big group of taps. Will experiment with it
next time I get time to play with reverb algorithms.

Have you implemented this in a reverb? If so, did it seem beneficial,
compared to "ear tuning"?

James Chandler Jr.

[y***@gpe-hkg.com]

Thanks Frederick! It's really sounds good! Waiting for your experiment
result.

Best Regards!

+---------------------------------------------+
| Shi-Yong (师勇) |
| R&D,GPE Limited,Shenzhen,China |
| Tel: [phone] ext. 380 |
| Fax: [phone] |
| mail to: ***@gpe-hkg.com |
+---------------------------------------------+





James Chandler Jr
Sent by: Subject: Re: [music-dsp] prime number's role in reverb
music-dsp-***@aulos.c
alarts.edu


03-07-22 09:05
Please respond to
music-dsp

Post by Jens Blomquist
----- Original Message -----

Post by James Chandler Jr
One could design a non-integral set of multiple tap ratios that "ought
to" be fat, but in practice the reverb might start fat, quickly evolve
to thin, and then diverge to fat again, giving a pretty unsatisfactory
tail characteristic.

If two comb filters c1 and c2 have delays d1 and d2, then they phase
from
fat to thin and back at the beat frequency 1/d1-1/d2. Similarly comb
filter
c3 will phase with c2 at the beat frequency 1/d3 - 1/d2. By choosing
1/d3-1/d2 = 1/d2-1/d1 (i.e. d3 = d1 d2 / (2*d1-d2)), these beat
frequencies
can be made equal. Then, in theory, an appropriate predelay before c3
can be
used to set the beats 180 degrees out of phase to make the density
remain
roughly constant.
This idea can be extended to any number of combs. In practice,
however, it
doesn't seem to really work all that well.

Hi Frederick

Thanks. Sounds promising, for algorithmically picking comb delay times
in order to "smooth out" a big group of taps. Will experiment with it
next time I get time to play with reverb algorithms.

Have you implemented this in a reverb? If so, did it seem beneficial,
compared to "ear tuning"?

James Chandler Jr.


dupswapdrop -- the music-dsp mailing list and website: subscription info,
FAQ, source code archive, list archive, book reviews, dsp links
http://shoko.calarts.edu/musicdsp/
http://ceait.calarts.edu/mailman/listinfo/music-dsp

[ccos]

hi,

any body know of free implementations of, or have advice on fast
sigmoid activation functions
for ANNs? possibly both bi and uni-polar, and preferably dealing
directly with 32 bit floats?
approximations, replacements, like table lookup versions are good, any
ideas are appreciated.

thanks,
_c

[Ross Bencina]

Post by ccos
any body know of free implementations of, or have advice on fast
sigmoid activation functions
for ANNs? possibly both bi and uni-polar, and preferably dealing
directly with 32 bit floats?
approximations, replacements, like table lookup versions are good, any
ideas are appreciated.

i guess it depends how accurate the sigmoid needs to be. it's worth looking
at the clipping functions at musicdsp.org You could combine them with a
low-order polynomial based sinusoidal approximation (google for taylor
series). see also my page on efficient sinusoidal approximations:
http://www.audiomulch.com/~rossb/code/sinusoids/

cheers

Ross.