Realtime convolution of two live inputs - using multiconvolve~

hello,
inspired by experiments I did with convolution using mungo´s c1 eurorack module I would like to try something similar with max and expand on it. what the c1 in live mode basically does is, instead of loading an ir from card into a buffer, It convolves two input signals in “realtime”. from my limited understanding I assume that this is done by capturing a small chunks of audio into a buffer and use this buffer as an impulse response. the buffer than gets filled/updated continuously - hence the changing ir over time.

so far I tried hirt´s lovely multiconvolve~object (mc). it sounds great but from what I can see its not meant for this kind of applications (continuously updating ir buffer) since everytime it receives the set message (to update the buffer reference) it creates an interruption on its audio output. I might overlooked or missed something here…
is this something that can be done with multiconvolve or is it rather a case for flucoma or framlib?
also if someone already worked with convolution in this way and likes to elaborate on it, I am very curious to hear/learn about it.
thanks, johannes

Yes, this is a job for framelib. The results of doing this depend very much on the window size and hop size you use. Note that the resulting levels can be very unpredictable, so proceed with care!

Indeed framelib would do a more elegant job, but I did that with 2 multiconvolve objects, the same way one “changes” delay size without pitch shifting - you change one instance while you are convolveing with the other. at a change of kernel, you swap the player and the one being edited. It works well and is quick do code.

thanks for the help guys.
@weefuzzy I will definitely look into frame lib in more detail…

@tremblap interesting, I actually already tried this or something similar, but didn’t found it work well at faster update rates/smaller buffersizes.
how did you patch the switch-instance-mechanism? for now I route the set message to one instance at the time and switch the destination to the other instance once the ir-buffer is filled again and simply crossfade between the outputs of both instances. is it also what you did?
here is my very basic patch, looking forward to any hints…


----------begin_max5_patcher----------
2775.3oc4as0iahjE94t+UfrlGRxzwacuf8orOrZz95n8snQsvlxtIACV.tS
mLZ5e6acAvfArKaJxjn0JwtcUfOmyWctVbp+796VrJ6EQwBu+o2G8t6t+796
tSOjZf6p99cK1E9x5jvB8ksHU7krUeZwCloJEuTpGNa2KK2KB+bR7tWqmbeX
45mhS29XtXcogFT7RvCdz.f5Ce8WPnk.u+n5VROrKNMQTpoE53fYGJqGkTMZ
bjgvq9z6w0TzbYkecuvPtEEwaSCSV7P6+JItnr4Su+Pcm+082qd6gogAIYgQ
qBS2dN4mQUxLW+N0+7xObP4G1W9QiJ+Z14ljwUZtWjW+S+br3KOGWDuJNIt7
qsYjrMaJDUxmdcEzRfRxV+YQTTd31h04YIIpKCTM25j30et7o7rCaep83hzv
UIhm5eClIdt+DFpTD+MQ6QWscWVT2QxxijhTqQBy2Vo9u3Cxq9PhPpZfZX+A
VBQbzRpT4Uu1g0Bb.crkPvYWBSC2oYuEQhcYEYGRiVJWCjzbQ+0XH3RJ4tTU
tzakWrW7hKBCHNdIFBC.XNjBwXFi+fGCdKJ139BMkcdMaoUbbZY8GtD.dmGD
wWVO7lrzxlEq+UdrxQx3HS.VgLXHSgBDvs3jC1nBuNKIKuqwU0avV+jscFLt
2fMROTJXpkTUay.kL3Mfdqy1sSjV1C9hy8dypCa1HxeqmhDdmymHLXIjRkvF
GfA7.DhHEt.+kHyKlO0GhCjiQMXI35zq.CnWgV3NUk0RC2RQtGzBqEBcIlyf
HoDgnX.E5eYqEpsgAIiasTYkT+eWasHh1Jd0BwGxWJEYLU5o.BIAn.jz.AcK
NKPCH9vK5r31CGJo9piACGPDILfbokSgDLP5TDPYCqDSANRbQ9G4lbouIoJ3
il.jsU56AESCF1IJJB2J5o.nsw+E3UhOHDc.8A9TbX1FfF2b31D9vxx7Cw2T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.ZqLHu7380JQKZVkih2JJJ6NVYnYeiNNRQ4WMfdqgNTuK
ZOVJ1sOQJEcu.o0QbQYwSYeon5BqUzZC.G2f111zscE1Y7y4R7D2hgq+rDAT
0jzd9NJ7+lHU7bX6oGvyouuQEG2ZaMaa7NV1Gi6x7T2l71bv3IhzKlZfIj5c
0gRp7OciXoV.t.VfLQXMat8.19isGg1iEP5YAiSBiXbDWsOMlZ0TyCa+aDmF
IdokGbm.VmQw6cuNM8MSATv.c3BLcFz2frqFimO0tyfjIwohWchwKWmgLFNE
vDMBXxsALOtwD+cCpNyunQO0nlNKtEg9+P3W7LXo3k84dEwou4MPD+8+xF3a
eGbIkC3AL7+PNzacgxK2nB6itLDCuZHN3mlPOFvvXBOKQdZ1K.ad9B8B7zTC
Vt2a.u+z09lXP3uOJl+5DcaRM6CEBMawfP3efhAYHucXBgZqF33VifQvDxhK
ocE4UGPYlSvwFSRCh.MZIyiMoKxFDMHXg94HavJL1n0Y16.WaI9+OYCVkZck
as4HaPzOYYC1jAiL4Eml6hApM6LMjBmgbWP+XjdX2tDnK5JKhKva0W89MQ9S
g4Qd+60eVjLQbE2NJDznP661vPAKtJj63N+oLQGY+bzhhZ9ggzhrC4qqW5p1
SDutBUjnnLNsYeP+3wxuTWnUqnWKSnTwriK7mStvVrP4I2CNWbAyVt.OmXA8
ZvhYiKv1xEj4jK.1xE8Uia5UP3eurlp.zwXMvrvZCPwgYs4zwBzVGKnYkKr1
I6b5XAdUNVlMtfdMXwvqHUCV+HBWndNPQOZdlWOp52k3UGJMAHa+LOupmMy1
jrUgIm7.UF5Q2b+QlyMOmYYNHRSyW898+66+O+9k6GNX.nW+g4Sb0CZl88rg
gMM6Ykj6AAfqsUvn39cFIB3ujqdEP..Lwmwv2dqycZqzhvTT.AB4LBGvN1Ss
i1XsvK0XsmVxxhNDjBfAL.2mfgXNh9fkCoYl4nqpLMFCOvo8TED98smpNyIw
P0tUmsmbM..QC.bfiNJF9NtmCGqqn1rOI7qu58gvCQwYpNBwKJ+vNUeBtLLd
i2GT+kERecO0U+w00sXCz1ObK5Kpau8SEeKJbcSuN8b3Qu6gGJydRjr+Q4Ez
H0sZKjVRNDpZ4X..hT0dpdACT8jIQqN.VJGvm6CpdgFZr1Gukv18Wv3sUD2T
fuonTB85.5AZgcbvPhYS5ptQM6WOpBccGABHvr8Qli.gq5nWBZw48f3tlpaD
.YaXovCMELgn7x5inLeeere.0mSjA8QNx7CO9Qgqpo5bGDMZufKJkofB6lDz
0AUlcvAGPVxB3HBgQHPFEnxOhG3HkIL5Z6qUmpIY5uxrzmyRdV7pFv9lHOaJ
3EwnFAwTGELChstWE6eLkvAX.GMPJUyGl9tWUDmdacnuo2kXXm5wBYOB9ygp
Fclzznem0zzWS2Mi0fvmVTbEN2qX3NaQxH0Ai6dQc2PodqsNmxvlKZWbz9LY
HxJYkoq9ji0sKfueyW5yofIwo9VvopJF7FqDGaID2BBwbAgz+HvKIRz4XY2N
R2YWW6stKqwt8KTfJUT8oUE4q0Izey0pApSwaqyT8vbtJ2tw4broBUL.tjPB
jEtKKZ2GIyYWsKAv.84ss+bSbwlxrwRC4.0JJyh0VjKnDwFYhxbAkP1fdXWP
InETh3BYBiswiVWy+twqjuXL.WVxY.FRz65D.5Cf.n9DmfwpgjEahP9TRP.k
i3mcSftJN+J8dL3QO+XP1owOVqc.mJkrfPANP2.YSzNjKzBQ1XCCcg2h5mM2
4oD6bdtoDndilXAKI.t7ebkxNknRXCUkFY+4lJeayhtKb8frQMV+3zPiFZCX
J1kxVxvbFBxY.LyGS3pCKtNp7.yMU9FL4EVFIXYPuCepboL3LKyAtcY1FKNv
Y0N45c8CBApOFVjTy45ThrRsYNxhDZS0CHWXZ.swKEwEIkeR.2y+TzmFkv1R
ooF0BZSNM7YQCAd05lmXTY18DHTWXQiE1LTXA7Z7hMMJM83fU3vDbPZ1yhSN
luJN9ji26IGs29Gq2wORumdbd0Om9wNFu2+W2++fwl8Qf
-----------end_max5_patcher-----------

Your patch should work with one caveat: you forgot to wait for 50 ms to change the kernel…

so the leftmost part of your [t b i i] should go into a [del 60] and then you should make sure you do not retrigger for (let’s say) 100 ms. that way you are changing the convolution kernel when it is silenced completely.

Try that and let me know if that works better?

1 Like

thanks for your reply Pierre,

I have tried it and its getting worse. it doesn’t sound like continues changing ir - specially with faster ir buffer update rates.
Actually I think that I already take care of that silence-before-switch-kernel part by first updating the kernel that we dont hear atm (via set message) and than crossfade over to that kernel in 50 ms. at least thats what I was trying to achieve :slight_smile:
my guess is that at faster ir buffer update rates the crossfade can’t mask the short kernel muting (after receiving the set message) anymore…

if you tweak the cross fade to make sure you have 2 running at the same time you will get something smooth

the order of things to do to correct your time flow

  1. crossfade to the new over 50ms with a 3db pan law (during that time both are processing)
  2. after the crossfade, change to the next kernel so you are ready for the next

that way you get ‘continuous’ - it still sounds ‘granular’ but it will always do… except if you do this more sophisticated version: you have many running (overlap of 4 or plus) and the changing procedure is crossfading the input! You can detect the output of a convolver to see if it reached the end of its convolution (outputing near silence like this.poly voice allocation) then you can change this voice’s kernel for the next time.

That last one is as continuous as you’ll get but will be more expensive because more convolvers are needed.

1 Like

thanks that improves it. (patch below) will look into the other approach later…
btw. is there a difference between the framlib convolver and multiconvolve~?


----------begin_max5_patcher----------
2985.3oc2bs0jZiiE94t+U3hZdX2L83RGc0ZeJ69195T6ao1pKCXn8DiME1z
oSlZ5e6qtX.av1H.YRxRMi6fD1R56b+ni7e93CSlV7VR4jf+QvmBd3g+7wGd
vzjtgGp+9CSVE+1rr3RyOaR91USS1L4IaWpuklmkTY5Cpabcb0rWRyW97ljY
U1mMQRCQOEPvl+vP5qXbHJ3+d3AUrsZ2SBW2psoputNw9XlL4ofISiyWNY+M
lN2LsJl9G+FkMY+LXS7pjpjMOmjGOMyb2HcW+0iOpu7jqK1jundv6dpUIuYV
NSTS0xjEoYUmAHVTjWkqlHla5etIMNaR+PDkiBIBACnDDVPPLtBhnLUab.yh
T+upMH5o.gACwjdQOnWzqSTCMowzsL8aleMnd79CvJV8V35j3Omkt58NQLb+
3BinWtbjgmIhbFNGZeq8xzk4Z7u4+JKsrZ+e6BaHS7GFjUDO2v5dghNLNSul
Elqrnyr96k12qTC9ZViSMSzCr+ull7kWSKSmllkV80VSjEKJSpWJFRHpwbOq
X1mSlOeS7xxYaJxxLxo08MKKc1mqdYSw1kuzrcqD8KmdC1Nd8zNrixNN6csN
c4ph4saoXyb0RpQKsnQ6ZLdyxZMlS9n5QrUodQQR1ul5fDhE3Plh40J2ZPAI
6xIg0LtGtscpVlmrpnrXa97PEQRM9SNkJCHOxJWELMHMH8R4jqgAkxsPB.RD
Q.LfP3bgR9FNCiMYPFakPbZd0t+zEaNi6Q.3CA.VLnhrKR0OVRz.CA3Fijhq
UFu0Ze5xYEYEaZK4UeAZ7rapJ.2oo.7UYJXVwpUI4UmfcoaB9aS2tXQxl+df
dHBtXEhfLDXLEnIPDjPhwTEboLXgse3QJqkDoZUV6mApWjD0AeB1i7IyTBkJ
mDBfNWiryKpzk8+yJpPGh+3IqSTCIpP8onRx7kIuesZJ.QnZISXJ0D.PkXI9
o.J9Zcdbmlh9cfDtpE9U4Qbmt60AS7H4qLNx29JuJorLdYxIL.FY7eAtT295
De5xxA8pUW1MvbUL+wUUa1l1bs+7WRmWYbbgp0p6qnDp8FToMv3TH3Xn.54W
sWnEqapjeQ7rV9.4FHACx73uvIpJVtLK4hMPPvMvHaHm5q2j4zlrHz6krylD
suydyOCNEER0eTVNwboR5htGlnHuFUI9NDUY4qKdO.hPgAnNwHx0fQJuuXZW
GLthwi7Y3lm7u5B33fG8AaYbZ96Wt3iYoSfCNiBHwka+4vx8vBc01rpzYuDm
mmj8ZrhTjW0ieXx6TtbVGnhWrrbQ77jAYh5EnXDVnfPjTNiCJa1DsPEkIuVq
RCva.xVLt1XUeX+5UwxjpB.NYSYZQdiwPo9e85FM+PiaQCR+gMJgnm12TZts
Ix9l1jnCu2d+GZMdiBSpT.x1MVoq236zMpeLp3q2jqLKpmJ1FUjq5ojgvnkJ
KWWaBxP+108ATVXMxwr+IBYTmKPMBgQGqiMF+F7RJjccRdZ9ZsNz7p3p5I+9
tmmrHVwM97whYc0+QVIa2YmpVdXxxMoyKx0ShVjBcy6FtOozyXDuXMWLleQd
75NtYESfBW5oyR0hba4z3MlDwXETv65rpnHqcW6uurjEU0cuNUIX1FEqJV2e
maRW9x.26zBUmqF5Ya5o74s41deVwTT8bY7qsQ6p3rrZw21O92hySWEWkTkZ
IAXz9NOIMQP6ddsidlqXxmkr2wMTSlA0OOc8NlnI6oxySWlTV0tspXaBhNzR
Y0Wsfdil1tKGZOWkrZclZUz9GnDORKqJeo3Kk0+vcLZMAfCIuuoPcScgsZeH
chs0K9qu2r4t78nGaqA+qhr4Mu2tLxZbPiDYURJMw50HJ89UW5jJydSqwolU
QVypOrybRsNpqDNa6FWa77W.UnbNAocEOLwhTTaBhNw1xkBX8BUzI+L.G6Rr
ynCGDeBGCHrkklm7d.514PH1sI4zTD0uiamy4sVoP3DQK3mFDplo4tiPn6DB
UE.g8hMvfprOm1ZpAr.pHLRE3XjLhwPBLkI1mMNx0K7sHqHtpW3CX8p6lDhk
bfx4J6HBLCQ+YGmw1vM+9fy7ezv4xYwYIAbpdqVTPtxardQc9sf5TL2h57PQ
DowGP0n0+7aA16GwECf3DNG.EbGI.J.76IhiTg9q.6g3xuI7tVObO3MLd3cz
Ob385rzJCd2KRS7f9DNJTfUbxjHJ2bUgyBaRygq27WGa2Zuoq3TDGBYQfjpP
b.iTpWjdDwMnWuHJxMcvVeJ46tdsbiCAOzVtxra6RaFCn4FxmmXV0jeTfH.6
UH5LAuQfKBkv2GQ2OLlQGWCvDlMJY5HGdLUbmBO96EdVmsg6EbxuSv4MIDWu
EY2IYX4EICCdDjrSUm7WuWThVuQht.SndXJv2LBbHAw5Xg6Isel0ft+tglxh
salsinUG+9SAsWOySJqRy2mu7OcfqN.Z8CWkNecgx.W8TASiLEwjMRElvJrY
Bv1I54EO0QtN0EmapCLSJHIDtcu+L6qi4aizb2fltL20otLXzPPrqH3XNKLj
Gmwhgoi1MukDYhvnVeeDZzXAAWIijQD+zYHwMp3XRDckFBiIRH9A.IzgY9cm
mf3J4.Ml7DtpfVGm33AExKwB2nMKfKwX0nwa5JVzAS79ZiGFuoF3xTSz+TC8
cdpcp4fi8QRXqtHLQmgWlfKwDj9J6ti0NwFvN2BhvPWzBB0oisO1rzPln29+
4OaK0gm0EtX5zsUVGdaVqKWzVxuLqXZb1Q6idW6X+iGlb9o9hTASnV8uG76+
me6e+6WZEFUWGzfDcRcAGQGiBLxmU9ssH+qW4pfeQ9nDoIHHD3RtfRjbUOQR
hNZ9nPg9ijhPDZDmSttRl1tGImT2zGeTJvDFVRAPvoBD+vYpn2CVAr+fUz5A
wPfjiDQThxyUL6IGaB14P6coZZs9TKjiSszBvcpN+zG+NcAIzyYw3r..TW2Y
dsXyi7n3V4h0Ywe88fOFucdZgtR.CzJ6J9bRRXb5hfOpK86fKtN7qW80o.AS
u1SiPOaWdSzPbMnQx2lGOaeMt9Z7As6waqJdIIa8ypevdlrFUCXyLtB5DUfP
.Fj1OfTWc8lbERQgpFhDQn5O3tZ6v4+6nBNDFBvyJlEm4T83JrUr.ks6p6Go
IkKv9iM6WOCKzkcv2.aMYXO3aT+bv25Jsa9nXp6APVFWkDf8NlP0hZQXFOJJ
hDIYQBpxnO9VNLPceDmIdDh58L.kTo7fFFvInqCpD1L0Iogbo.SobJE3Lj1+
HgzqZpI2ANIawwWj+ZQ1qIua.ruoTf6uWw.03E0xFAD1X3BIoSWcLm7ThTEb
.tCWkFOL8CuqGbl+34pK5+ZjTLBHHl7+ArZr6.mFyubZleS6MWwhvGGecMNe
Rb0rV6QReI3oUL0GEY7IDWWGZnUt1NS9ItoQR3v.wAOLP.3vHI5FLgaZjot.
l3VI6437hP4xvniO8pbZ3sBJ3HGlZTjGfeF2gQhf8vHIcYfjiAg14QF5iNyM
4mQPMErQTz9u3a4aWH6fOj5LhtmSGFvFCpgaCMYHwNNIrYEERvRcrRlWtF3H
Cox7MeScLxJ3yIUhGZlW+lwRmdKJUxvRNUDgUQfFoKSO61SeZe9PF+bHd6cI
1a5XcQ8ByGFrNB26QJm3iQxEKVTerlHDWzGzVJssSRpObNRHAoj.TSJLQPDB
Pf4jrRH5ljxHLNhQkRl.KFLyiWzL+BEx678XyAO6ts4iybGvsNRNLPROvafc
QxB7ggaL8Bb5ra0dLJXRunxcIJRn9OglEjQ09tiqCd3z9t04sKjBenP.6Byk
Y23v8BPHrsjV35ZTWJH3HjTI5JD25bCcyDONUFJO40IghbIGfTJ8KozEO0QC
xAJLI4E.yFY18RR2mu8YvIViwvxK3hWuXev9CtnIh5kXEYtJncyiDw0QBtGw
+BCFDp8c2AX7+cOe9H3+Kb61Apmc2fxCa5aN5McgdFeza3hid6Vb5a1h9eqV
b7azBSIKz2axhG+qG+e.G319K
-----------end_max5_patcher-----------

the former is more flexible - you deal with frames and can implement, at the cost of coding time, quite much more powerful algo, including multirate processes and stuff. Not for beginners, but incredibly powerful. The 2nd is a blackbox which deals with all the partition convolution time alignement. In other words, it does one thing (r.t.convo to a fixed kernel) very optimally.

In my tunes, I use the 2nd. I will soon reembark on FL for other things than convolution…

Ah great. Thats good to know pierre.
did you also implemented what you, alex and diemo described in your paper
Rich Contacts: Corpus-Based Convolution of Contact Interaction Sound for Enhanced Musical Expression
by using multiconv/flucoma? or was it just prototyped in ftm/gabor/muliconv~ back than?
i would be curious how this basically works… and see the patch. is there one arround?

indeed just that.

The patch is in the repo of the article but might not work anymore! It was research code, after all :slight_smile:

http://eprints.hud.ac.uk/id/eprint/20131/

1 Like