# Impulse responses with NMF

Hello !

In the piece performed the other day, I shortly used decomposed impulse responses with nmf.
What would it mean for you in terms of signal or architectural reverberation ?
Could it be some kind of primitive decomposition of the mesh of a hall ?
Another step into finite elements decomposition ?

In Rhino, there are tools for building meshes then convex hulls. I am looking for a link with one of those processes that would eventually describe what nmf is doing to an IR. IR are all about delays/distances from one sweet central point. There is something to dig thereā¦

Just trying to get an idea of what it means, and if it has a meaning at all.

Thatās a fascinating question. Intuitively, it seems perfectly plausible to treat the impulse response at a point in enclosed space as a straightforward additive superposition of the paths taken between surfaces.

But because NMF is unsupervised, thereās nothing guaranteeing that the decomposition would be physically plausible, without coming up with some additional constraints. Iām not sure what form those would take.

Something like finite elements methods, to the very limited extent I understand them, are already starting with a model that is meant to be physically plausible, and these methods offer a tractable way of ārunningā the model numerically, yes? So, I guess, if one could formulate constraints on NMF that somehow represented the modelling assumptions of some architectural acoustics thingy, then, uh, maybe? #outofmydepth

one idea I just had is to try starting with the DC seed I use in the examples of āupdating basesā. Imagine if you make a low, mid and high pass DC filter to seed, you might get some sort of separation that you can vaguely have an idea aboutā¦ this is untested on complex signal, but works for drums and sine waves

To add, seeding late and early reflections as dictionaries to (miseraby) split your IRās up! Perhaps something wacky and perverse arises.

In fact, I began to try this in my patch hereā¦ but the result was well, crap. Actually, NMF straight on IRās is pretty good at splitting up earlies and lates and that alone is quite sonically rich without doing much else. It wasnāt for nothing though - this was actually a good chance to experiment with blocking as well as bufcompose.

----------begin_max5_patcher----------
5904.3oc6c01iipjc9yy7qnVmUJYS5AQ8NjHspWsIQYj16UWM2qT9vNQiv1X
2rWL30fmY5rZme6odArAa.WXC3tmT9CcaCXSUOmScpScNm5g+1aeyr4oeMLa
F3eE7mAu4M+s29l2nNj7.uo3yuY1lfutHNHScYyVjtYSXR9rGzmKO7q4pi+i
ggKy.4of4gf.vlzjTvpn3vxqKY+lnj3vb0OBrxWdQZb5NcCv0w2m4x8HXHli
nD+GDGBJeiKr7DMdHniK3+o32LZop4jN+u7NJt7tuJMIeUvhP4ovGaQo6yKa
RtEGcaP9hmhRV+ocgKx0MKnOww8A.mRcnh6EkI+DxUdOkem+9aeq7OObi32u
7THX99UqB2ABxA4hOElrr.OWFJ9VaSyBW5.9P31XQOQbEQYfuDk+D34z86.4
gY4frz8IK+Mk+zwQIhu29D0uOoK4PELCyFRLi4ij+CRwpOcsXVR3WDssyfLM
b8Mv6+.A7iQBEo28mRS29e7tev4KAedlgcXHhLqwdGp3n5Ck+71PcWa1p3z.
g7Sb+CRVO6fhWinfTigCgJPfSUJNnAVw4OrHeePb7yO.hCyx.Q4g6BxiRSx.
Yggax.KRSiC2YJbfwy5qrlCQNbQ+jpE1HWxMM.oEg81hQGhdXRRTdi8G2F5O
vY8W2UJ0XTU2vy8nPq5WIbWQWnnOHFuHr18YQ6S.8Uv12LKX61JG9MU9JxN9
eQa7y6gCGJJQeH3gCsK7yQkee2CGMXmngmKZ062ozLm8UVolr7mIcY3tj8Qp
eI8AEhfhljBrSB1DlssX.tRlTd5iPA0Wo4pknLhBKHdtGACgvbcb5heMbY0F
2rzsgIQIa2ElIzXUZi0N8xvUA6iy+jzDSVz+qpE.kJOMb9RSPMdRYePOHXWT
fZdbXE6GBUXw.v5m5v2KNbUdwo2JTqOAEyS219I2Es9oN9tySEmbSW+1pyj8
o8I5y9IgNQ9mxB9bczNWXZoXDY8e9uFjDsIHOLORKBPtGNYXRfni9T1hcoww
05u5y74FNyRgN9hvuDsL+I0Mppxf3xi1VpDM6fTdYzZwjd0OVdv5r5GIK+YM
MYjtUC0sO0Uel9pMigT0XdnudJLJr9TXuoz7dg8kwCtDyw6bzMH.ruvEziMg
3EBU32i68BujJ0NBcr9iTX5TpXUNCC8dATgA6he9JQJxThTX8+f2MUpEOEjf
bjS0kcEi+n7IArDiyU3i1REDAumnE75QKuozZ0KBvRnZ8Tv1vq.rlHKVkpVE
yEh7typVWKZMkVsLDrN5AtLnEEtdUbkkKaYlz2zkeR6G9mBxy2EMeet9hqtN
rd4unXwKyChK7F7vpIZvcx2drEOPqa+mhCdVEmmR2ZhjMYGv6S.AfUAIKhVs
OF7kzcwKePFiGv5vh3BImwBDjrTMIOXW3pXAhnVwe5J0UDreYTpXcxf7ujBx
B2JiHPHPcWRDMkrFiPDxvHDf38dE0LhuJBAbgZf3eXlZ8jX7UFgf7z0quPjE
qpv6O6PKamPBmKb6WuNjps558EXaw9IRhUcDwGlKV2SwpHdgHk+cHCExxfEe
qxRFpi.nFP.Z+kXE8COW0.W5sE5p1BoiXDvYd1i.OFmltE.MV91R76vsICyh
VmHCVP02cwP4cPvx7zKHxcTBl2eTdEQqdVOzVNF+3ncoOpxQ25kLJzDSkeLZ
GH8KIkAGyA7KOElHGpOWL1VeQAfLQeINDjIT8CSKixr3L6yD2iLwETtLXPXb
V3WDlBCc5zFQ0H67oZWgmoRsq1HBzWMML1i5fn2fUjMgYYAqCOSDjkte2hvi
qK3AgA5rbgib6xkS7JZh.o1xCfhKrbkVmdcP80Ydz3aXjKwcV+rOYjNLwiqz
gYpEoxF3Q0qh2GszQnjULj9afGmKCQnns.bAO9W2GtW31B3w0AhInbcvhWfG
kPWThRepVjQLNr07dZDPKZp3TiQHGUubUBGdSi9aS0qRr6o00dtnIdO5XnoP
0l4ft2V+U3mVtvaLSMMPIC3L1FaYuvtx3LkcNXNXQrvlh4f.pmokpTiVeaLo
GivDcHED+S9x2mh84TFmOB51u+CXiUlIjQwrmNnbHU5cFg9m49iQviQ+CKVF
ORIHoHBE45wI2bus0jw8wYhUisS.DB+293Ly8UCMJRVo+opWDNkv8gPjv0Ul
MyckYtiqbZkC41L2YybmMycWW32JWTUFvQ71Vi6Fp43tgtxntUy2z1LDhHdh
ECIr+I7cxmib4LgoeeOGtu7EAREyEP7DqHwGO0As727tNhQYKXErcrB1EVUM
5LsBUPWGHUhTtdTervidrXpSeeGlxEHN1G664SYmO04.gU0cfnNXAcklpEhP
2ZYKo1ZGgvdhl7qDMub.w4JsILRuHA0bLX5DBWb47a9behOpU3B0SzhNVnkm
uFrTq8VGc7AGr5L4AKW9buGHhFUyV9mazh5gTtxAEtvxwhwATOg+ciSho5R0
5Xbb9s0Cji4Xmq+XoJ0nMLBUuLNwh08QXeBhxkN.58pD57FKnSG+loDppksB
iSSmWMkmNhCxXLGoBjPdb8bhdtbWByy2EedzQFe6VhkhrUVIz5ns1WXjCGKE
ILyWulypVufXTC9bglZetp3o57fr9N56pq8.ir42Ht0jqp54KmRXqgHWKiKs
JP0Pm5Ao9P8U8n.fjF6jEwWk.c2e+QZQClzgFLtKoQ6Q3tqoUHRICigoLp3O
ByhDepCUIaXtLlqOlHipDkO09uba1AvilSvMtzqWH1A5iB8w5fquZz39JKXS
lFcCNX1n9LjqKsc+oS1HyHP7jUzukP2U4RtVukN406zuu29M5OarcJR.FPox
iG2m56wfblbzN1gd5TZSIPkCX8udnGUn5bKhb94VDQStEwsgg+ZsJYsUbC2F
tcsAHPa05hwTA0jtDrAvC5xtenG7ZQOxnhdMn38hA6.eT5WtXkbUqDuCoNqG
CbcGu0Fe9vVYQ2bF5QPmlZslRuVcfp0zr0Vp1N4q2TJ2ZIsasm5sKm9s1RAW
szvYPp35JcbvxxgE6UGAaNmbFjWNSxM2EyOmg4nqi7zYVt5tP95tXN6tXd6t
Pt6tb96tXN7LHOdljKu9jOuNxo2EyqW241q6760cN95LOeskquly2WK47yn7
90bt+N0BxoVwO67mXMWaAslogNLRelu5M7EOuRG61W.cdUz4J3veqMPRYIpn
XROEQTMnjkge8D6XUlpq9zc2Jfc1zecmBhN8quq7m18peLL3TW.006Illx7v
8C.OrddkyWhUyWTBpU29LOp2xQU2rVxEzKlPK5yYxBvrwUz2KMaH+B513KIL
5dk7cKY7UynPXnWlRlshl6tfXkvPHgVEFH8yHqtLpZAwVohgqG5Ew6WIC6el
jRE7N8V2VPE6ojjbGkjThLA6DcMITOVhuLGiAOeLFbrFiQo2EICEQekOHC18
PspsJuttA7hTgn6wE9HANu8WIJ+UtVr9ZuJUgqs8ojqF2BUSN0vUmtaoNHCv
XeZv+tpXDz8tytocnc8Nv.DQ2SmGq04vpuG68lvXzx00xDrHBs9r1lMqsYxZ
XVrA.4NK5IFROAcjCO2diMLFYbQmFYtfSArlFMWA1NaTr2YChacVFuytxymd
rSg4301Zuk4NJsLNzvVF47KbSzxsoQI4EBPL267rvII9tySgIz2w+rZl47q7
EIPHqDigBHfdNb5kuvoTUk5YHNTBXiRqfXpzPUr0U8XXPaFxJfxrlgLAicqU
HK+qyD1EDiawtJ+jOgoEaQfy+Z8nHx5cel0KnezZFXsb8hMCp2Tayz3lFZpm
mAi6q95nzLPlZJkOhMBnoy4VN473zJ7nlZO8LkpQVM17llBJmzYgT07hQsM5
zCaj9.atSKr4ZrAfQUo23VAcpMd2i1FYxEeXSabX7XJ9PlZ.Wsvywqc352yE
mLRMCudtzfwoYX7xaaWl7cB6.N3DrAbTnVDOpCwuxKB1RvFVB1vRvFVB1vRv
FVB130CAafIPY.MQPe9fwXDSD+ZPXSN+Z3SbkOssX9TrkeMr7qgkeMr7qgke
Mr7qgkeMr7qgkeMr7qgkeMr7qgkeMr7qgkeMZZa5a4WCK+Zb0nmkeM5M1Y4W
CK+ZX4WiZWkkeMr7qgkeMtz9+zxuFV90vxuFV90vxuF26wXV903UvfLK+ZX4
WCK+Zbw8VukeMr7qgkeMr7qgkeM5LhtV90vxuFV90vxuFV90vxuFWWqvxuFV
90vxuFV90vxuFV903BC+s7qgkeMr7qgkeMr7qgkeMtV90vcT3WCjqCRsqcJ3
WCOlkeMr7qgkeMr7qgkeMr7qwqG90ffcYNXlGjQl8xjFMja.jodG666KlmmB
c4DjkGMr7ngkGMr7ngkGMr7ngkGMr7ngkGMr7ngkGMr7ngkGMr7ngkGMZZ63
a4QCKOZb0nmkGM5M1Y4QCKOZX4QiZWkkGMr7ngkGMtz97zxiFVdzvxiFVdzv
xiF26wXVdz3UvfLKOZX4QCKOZbw8PukGMr7ngkGMr7ngkGM5LhtVdzvxiFVd
zvxiFVdzvxiFWWqvxiFVdzvxiFVdzvxiFVdz3BC+s7ngkGMr7ngkGMr7nw+u
gGMpuQ5Nlqq2+AXibmApAty.AGCty.4wJ3NCJhPQth0uc9VedXXMj46WsJb2
2NroZLl0Pfjl64n154GxflJkZcACdtt5Eq36Sw9bJiyaASz4JcrvjC6KCyAE
7nAJLeOCAEnq6XfJGRAbxlUGyM+6+.A73tvrmSxeR9AmkgE63kkfGOVrs.D3
w4AYgYx567vG.BrcS3RG8Gdb0p7LA5H525TBCnPT0bC2XNgMPnzB+1faSnzd
Ne6ZPKCp2XV9mr+r5I5WFJzSg++v9kQpcKzyBfGThxpCYr5oaKiYcau2oCXK
u6c+dS6b99c22pUOxdNczk806NacWtPZyGVCHaAAE5vlN8pOsmytlEsNIHty
gqZYKkqFkxfV9npjOpH5Q5dtV9nxxGUV9n5Z2Nfv9l1U1Hta.gDcp.J3NBzj
SHI+VHf12sRp23QaRd5MAj1AYLZ54cquAfN8FOt1s83IyFdITQSHg3Ie64I8
p+a8FTHW41wt.TNcIWWBbvZvgL0fy+RuQFt23ptn4OKjlAxFI9XqC.4et+.B
kTFxQCs8zjaARQTVb83iBRcU0Vnu63pbnowwWBEP30AOnwzSVw7MzaEhP28Z
rzGe0kXYIuURGZkjQoNJ8MM4ediY0a369RnLz7Ls9uJyU73zJLtDRNuFRNoH
zf95AiPrdNsBGfjebrZ7lVsDSeo.XbSS50kQ3ZYk6Q7qVGe5OUf4ZRyp.wmv
JPlXXMqMlovlaZhz8Gyxafa5fZ9XViLb1KhVgoUU.aLMvwLU4jyG6VgwlCFs
VwsKP9NoJOxCl2Y1zdRbAKRi0Yk4OqBMck+TMoehKrvN8r+XYFIkg6s5GQGy
8jboW09gkzdH2ifgXNhR7evvCUsITckpGxMVGLTroY7V5A9CmPTGM5EtviSI
uA3ooCBU1ynUxwxrco6EtZpZk3Coa7TDFSQ9DHjyHbWpm2CFdH3vlNxkAK9l
vbAxzjQxu1jmWTX.zaqVUtXcA7ygREg9TO.Lud2iPDpl3HTA0AB0Q3w8J6Si
xXy2+An9YAh7sniuEqdaSCkX76vPohBqfg0OLfTw5f05HoATg4W1Esdc3Ncc
jHKvh+ofE46ChieFjuKHIKRUnEeIce7Rv7PP19shKNIZQHPXiNDjml53376t
oRxgcEkjAjpWxASGF.t7eX7.OX5mCy2uUiL5h.KCDl6.Bhc.uGDrArOSRpE+
Wu+C+BHZy18wYgfcgYBa+x5WZd3hf8hirIH4Yv1vzswBHSLoIH+ovMffjkx2
7LPxBgJ37e628wjOl7eGJ94yRAqCyAAfMQeEjtBnxbelBve.rZW5FveceTtP
XrQHmDh.vOmC9o.w8GLOZsSixBeSkEndKKnZC+XkDAVDttgtZQJKBuhAt.no
V0H7tqAu5SEw7XHNg5564h8QOz7gH52BopSQqa54FpoORAoDVjkVWxXVMiEF
C6APx5APdvgEWnvAF0D3rZGCp.xRTUAk3ACHQE7IxDAjvdBjzd.jbZM8OIpQ
O3DDR5TTiGZf.xhJeaXvw1rx9uWVhnBCh6Bk0P5+XFPLuYJH+KoUp1OSwWr2
U5YVAWfNVE3WaEXtvRt1QsLUc0.ZstWq8Hb4hNnhGzmiiEHTwiVpaz4UwJEy
M2uTjW+8GC1Yw71sQX8rYEOzKN4orx31QYSYGE45e25njoriVjbjAoetYebd
TVbjLrpE2w4qqYvtpcXBD8fgGpp0lKBen9CekFHV+TZV97.cPgKqhuKvQmyz
cXyV5U5tnFJCzFGoQUg2tPv.0piP5qZIC76BIChU0F32GRF2uKjLdnAbHSO2
YO3F2AOpUjIdG3wfE4QetvQF4AEet5dARutsyIjy5qSka593f28t3YzWVW+1
UPEdUH9YqT.tP2IZOYgZSxgZRxgNUxgFVIGqGRtwXcj8SxU3lzAIGdTjbUCu
gD8gluGCe0EdCg8Jb0WDg.ymOtKQWahxbTE8ZCU8HNrpu3pmsJvwNBR8RW0+
UW.jtGppndpp59ZCUuKppv9op58ZKDc2CMUX+zT4uxvz6hdJF7wY+b99kQo.
jyWB97gGKUWDdou1lxhRpVHaim8TEhVjereHMIsevJ40l40BX0COxC9Ev5O8
Tz1sWCnhekYJn.S4zojIMfsspMXSqZCd5p1fC5p1P8XQaigDqeKZqHaECzps
kRp5A39on5UUe8.ZA0A4gq5YpQ7UKYot.7yJRLrvtg177Iuokx.C4aRXtptU
6gmpNR5I8nTVdKU21FMRLJHtptv4EzIg6wZEpTwLbcvhmqgAkst4AK900pZf
o5QqTTLrCah7YQqSRE213nE+5gRe96WINy8lk3vQRhS4JboHwKV48vHuewO.
q+UfJTWxy5Z4ln+.iTAJZt+N+vVxBU6HlknPIQOUscIbJtf+g3GAuSRtpRi6
3epPsU6R2lt6PRGcniNJ6BuVX1iUAmQXKN2MNitRblfPUwYjEm6bMw82pQgK
wE5yLtJc9LF86EbVcQ0233ZX+zcrVA3eNgDWa+x05SKqiaSsyDvFempu2fa6
VAcGh6Us8hXW8J3Mdmp+vE4RLj9s0o7M4VQFf6juIRp5OW8NYqOOA2Z342Z2
a5VyM4V6W6QUvY6fbMKNRY5bDCUO+QYsW7il1zXbCzmOA5NoowvGZLRVxcnZ
YDWC.spOhPt9QZlbmPcAA5dMRWc8U9v4JRnwukB4FzTYEOJ1cO7ogVo2r1Jy
31JB50Qa81rMfLQAXHr+gLXzFZPlmxjYOPdC0c5h8IzPbmXlzmXC0c5h8I3P
bmnl4lzffeTSzygCT+xDUcHrko5cmjaNdL7y.RLRhRFDIJAaz8ZPjnDSzdHC
iS1Pi5VjA4dYz7TD7PbuPl0uNGC0qF6DNIWdaNgKxOgGxOmCxam+wOk6wU7C
RabN9a+6u8+Cmdn+Z
-----------end_max5_patcher-----------
2 Likes

Looking forward to play with this later today!

I have updated it to have some more guidance on how the patch works (or doesnāt ;))

1 Like

Interesting idea. Will give it a proper play but also makes me thinking of doing stuff like seeding it with amp-sum-style IRs then feeding it reverb ones etcā¦

Hello

So I try to get your #4 in there and I donāt get it. There was a typo in the order of the centroid but even that sorted, you expect to take reverb as nmf template? You are also adding 3 ER and 3 LR together (which is like 3 reverbā¦) What I would do instead is to convolve them separately, like in this version:

----------begin_max5_patcher----------
5254.3oc6cs0iiiak94p+UvwaB1MItMDIE0kcABpfr6hz.YBFzy.rOjNngrs
rKkQVxqjb0csAo+su7ljkrknntY6xS0MPUpjjMI+NGd34FO7e7tGlsL9q9oy
.+6f+J3gG9Gu6gG32hciGj+8Cy1480Ugdo7Wa1N+zTus9ylKdVl+Wy32O0OC
.o+22KI7kEoG1MGHtEBD5k4ytS9mI5vtfnP+L9WHRdy8dYqdJHZ6mS7WkI5P
VF3E1j4.SW2EzeAcQKLlCPzeB9axOUvZdiGu7u+dBpz2e7gr7F.JuaZ1Kg97
2N+8DuT1K68Es2rYf+F6I+y28N1OlqIhrILNpgAGr4AmsoCaz3XPX+hXnZnY
5l+suINYmG+avp1A6QzLwamelexm8i7VJF2Fc.HlClszKZa+.jH+uP60mwg7
a+VWY.rQVbF.GrIGd3rAMR+MFL8OMXajW3kjKvx0Pet.yeYxEXY4nOWfoyqQ
t.DQet.7uP4BjKFnEW.lbU4BZXPu6PXVvp3nmiCe1+arkF6LiBlyaXaAY+BB
UxofMmojuPePX9vfilTY3CeD0Yt.g7RSnU6yTv3qjBAMP9C5KKOxlzNKOZ3r
7AQYi5.NceB8qbC3Ce7Wutyb5V7o4XGGACuJBM5FiP+6.vtRpMIXFwkPbamR
CctdT5Uw614S+vmNj+K99qSAYwfk9.Ovt3nXvlfPekT8MwTlCuU79DtYnA5x
4DrIDtQ.DAqfQCJA1.qfQoN7p3v3Dw2swBWWKCam4zqfHhoaoKfkZgRP5HBZ
+zS9fkG1rwOA3kAxn+kezZIHt1m9o1Gm5udA3i96Co3D8MBRAeIH6IvKwGR.
Y9oYfz3CQq+t7u5vfH5m6PTVYLcD.eKgoWRv2ztAAtVpA+gghMLYS.feiJjw
D7WnqtF99+bb79+q2+8K9h2ycUpCjOwyFJVd01VkTGHZ3quRUWjpA2Tnf0dI
mUJHHJJHqVfvnMfvhXJzKU4xrPso5kaO+D4fRNpnrjTAFOS6wAwQkHSOLya+
9R29gReDFT72EyksmWbqfHwsvE2Jw+4f7OOo3tdIzQcFcHeHQz+9pUNEk80D
u1OI5P.uqHtIkn7t7uvbfRHvFZHllXJzNwojzIJ8aaX7pe1ecIPgxJr2OJHZ
eheJUxfWlryU7309a7nZK9Y1zzzf+Od+CxlCVyyymFW6Cinp9yGb+gj.llbx
WYaRvZpPZZmnBTytcdyQYBHhI8kGL72HxaeMeXJy.EVZ3gozA4gzkdILJgzH
DT9CyhiCq9nhOWn+lL4i2S4jOAEyh227CSB19jhO6xX5C2o56l+jzOeHR7zO
Smyk84Tumqh1YdggxIgU+5+pWT.0.M+r.AI.YT7PggXOktJINLrx3U7jmq4I
qoLwq7+Rv5rm3MTYlA5qGrOmIZVAUdcvV5ZDUuWl21zp24rInzacXobR5my7
2sm4.upuPE2FVdFYYwUUtuJwV0KQO8.cYy0KV5k5mVxXoFEn2jrLgUBPWgPc
B7TgYMZ2Pyx1a.2zVF+C4x4khUlNPjtr3hi5RTnXZOAQDRpGfQafHzw5NBEy
8f8PwOX9xEjVYBI2QvWQLAFL9I7+QMN9nQmebOfeqdxKBsfsDZZ+QP5LRNnI
D.BQsJGDRruyvP3PwPobuN.gN2ergO4s2e3rgxkiQNshg2UhBkrgCCCkrgc.
Bu3RCOZw.ymDRUEkuYtcTyX5Ru9yB6F9rWVVRvxCYhWtrggcR+VpwVK8BkZu
VX8iJkcOpQ76NNHFIW77Cgduv8rStNXArQwBvGh.dfMdQqB1bHD7k3jv0yYd
0Ar0W5IH1Rl.un0bcO.I9aBofD8ylBh2veCuCqChoV2Cx9RLH0eueB6E4sRD
sqjVqOgPc0MrltbWPaYy8OIV5UVbCNh0dJc.TV71ss3RQENQ2xVLMwL+m05S
GW0tVcfgRarcv9ZuUMGFIchklMWcdBQoetHWAu5QM47EgO8PfGCii22fK00f
vyRh.tXRkdwB1fy7vCHXYCxudMIQ4OxdifMuHjKvDPHDUTiDBl+J3hIX2MhZ
82G93BvO8jeDSJwRpXApzirXpfnTJxE5CRoL19w4tjl9jCozVHk9B4NA.3Gl
5+EpfU+Eip3EnqvwxlJEu.mRFQ5JOYz0W5ISlb.zhzEh4fibyn6n3LvRvpPJ
OTeG5HQ.4gP5un+CpRRBZvdLmO9oSrDc4w0i4eZFUejDus9zIJeZVewCLUPC
DgJbCMG35rv1k8O5k31LAEMpFfxVA3hY5928dEVp2FdAMV.o3Ejsvfq6BK4J
jmkfGMoqSyqNzE75ngGiDbUMktphWXSpAEVXCGhaegMnMmgBKzYGhZk8xdTg
qKGV4hoScHlPW2JSPpSo5NChNtBLz9XpxgIMCgjaXHToi0Vu9k9xm4VHLi3f
DpuhXYWXqNTCcqKOSESGUqgjLliHA+JprHgF0CW7lIov..nQ6tjzv8U5T1wC
83yO6Jr4bCCaU7PWm8psH0Gcr4KRZ3Tm0zmBGNZxDovucW0kQUHWiZDydV5.
lFeHYUuwTrkKm6hIhChQGUXC05JpvWmqGTRO2kdo8cpYYbqjdthkTUAaib.+
tnJ5tI7Pv5EKOrQltHeC7H0HnrsdAQ.3BweDDwMq9XFQ7HcLxjFxRQmGWxrI
mhj.itqMCr8kZLYxIsYaHPpBlr3vZ2lRMlJ3hwcmbj68oh.f8JSbPYy25l3.
r8qSwAcgm9XZpzUlZbuWxKmZHYog1hLH0UAkv5Ngkl444vYCF4HBdWhyENu9
xAtKFd86GjdlHJNAkS2gXpQmxU2Tpso6ssWHTxbYMbVK1N.HWDIB0dJO9pEs
166+yURQpFwNbKNtAwyS.DzdgCiACqmUMNiqyGDxxt7fGZXf2QFuNielulwO
vmXZsSsQqb5q7oYiv7WV3kyk4YpwDXiwVImS6oGiOWUvpw3z0Tr5N4iWWL6Z
HtcMG6t1ieWSwvqRb7NIVdVFxMqri3WxzUC6TkPTeD8zHpd5DYuVitmlQ3SQ
T9zKResDsuVi3WqQ8qkH+0dz+ZMBfZDEPchDXWhFnhHB1ZTAUGYP0QGTcDBU
FkvlhTX8QKrYoJsG0P9K0Do9rHJdpbiSkee1yOQNtPHWEABJDO2rZB7YFhMV
Ht3mUllbp160zlmmDPs.lrmjmXU08rfn09e8D4ZkV9q5RfCEJOaIQ0A1PoZ+
s.zhDVu93ZnJRspsTpEftMmYc8f1BmBvUW6afGEdUoRBt+nXqBTdqWv7J.co
ufmSAFM3VfALavkuZgoERGhDztk4C39SlT4QfaAZ1dJvk3ExISTZ2FeOltJo
UodGoNbJoLhMvpd1gd8FVfARoqWX5bZS2jaKG.MlXtfWOyv042xyowl+hkFe
97R34yKgS77RhnrL0sIlDxaSLqYhIT8DSiq1DSIQVLyjHxk0VmYVI302EDYN
lobBig1ZazAs5f1sfjvAntwYM7fU8qpcbk1YSMQD3fD68TSJjQ0QzukJgANG
3JErhRuKV7t8hIou8ONcS6dHekrZd63j0BGh.uk67Ftp67FW3NuYmP9qC2At
C8Qz0nGxVltS8P3ktGZzIlz7wiRgWuqNe001tv7TGU1KOdndWYpx8Gmr6LqN
tFAGSe5BeMtnmZWMaKRhKXSK7Uc635Lg9YtgE4pYAtQ.8NycP8zS87x8WG.P
EAlr875pWHTsan4SAs5lrWB5NaRtyYywabQHmydyyW8TIAc55aM2yLljdlMT
RxhQbzbLlCFSRuvTWjlm02kUDXT6FrLrRutAKLkpo3rLiKmzJKZiPqxkvw7+
BKBQ0QFAnv12oYDZ0Ifdx5FXAUr0tAw4RKYS6tF5RuZ.F2UtyIoafzUnn8D1
If5txX9RnSSuvgnqzyyXplX1X86ZbnDdIYj44JiV8MxkG1L6BrYbYgMCsE.L
oL8Z2KHWZg2cnuYdwIeXc6bX7TR9P5J.mah2z0OLb6nIDST2voiJ4OMcCsMB
sYZx8ao8pk5NtxC2JEkKDGr3TsA4XwKWHrMdpq5BwD7VtfieVs9ryGHVXqy.
Dng5h1DDeKfHMUpl9CxRujnb8WoBv0UvQ3SLSg4ZHQYRBAa.ULJcPQUJSF+y
dYwSYETpwyFhu++FfJkYu.9VsY0SfO7w9ACEEaoVfA7MEL792+665v0UrCgE
CWIsu7YuPkrc0IOwDKCBttWih2VQQGqqElNA8kHN3Erfpl66NhGwSuctD7VE
T5sJnzaUPo11RY8unGaJqAkl4hzTuSTrd8V8e9UP.o2k8GGQoDTn0GtUXx4U
5lR96d+2.vECEkDkEOrUqnz3t45pbR.N8XEyN1uMTnBKfJyVgpwsNYWacfc5
greWuwKYgEWT8sZuHkY67Zly521aXxQV4SQVZhS1+xDmxOlXzElrdMCSEUni
9mOLxC4Dwxe12sknGQw89eYvmnSxSPc6V0lxbTqXFsTmuuExKM4DObyIhcES
oMtWRGsAAZPwYbf1nFZBTRuK.G5lIO9xKtjDMQN78Q164pavrblxrQv03VHI
pbzM6kxi84zzKzNkHNOmHNIEpftxJYBVrlrTkOHdxROJsi9+kOz1Z20XpvoE
tlm.ZltkSGMweIwbQovRh3Wv7d0TybvZJCIqstAF1cJCWustSpsmxb9v15ln
WnaTxslRAbV5xbZaO08BsEGLY8hgSPteyZgLukJi1H84RAxy9i4Qlk4e6x+I
5jPich5xTU7XalcGQ0K.leRRU7YnMwp3PQLu9q.iEXBx0DNu1qZHFt1D0mJS
IwGnZ8wee7BCMNjtX1fVoS45ZYX6Lu1qJ2oNObhOc53yXdoeT9yVqt9yqTOL
o+byO7Qn335fcI53kX9kpldJyXDKr3Hzg6oDqFFs1i7rraqIKMwT8SIAa25m
HxNGVRp7u4sJ6fWX3KfrDunz.dxp7k3CgqAK8AoG1Se4nfU9.1AfFHKNdwhE
+lQ4bPCCIBiMrDdfvleXK1vwflE9ZLE7G8yNrWfUhT8JE3ms.3Et.7Af2NvA
1gHG3O8gO9Sffc6ODlxNzJSoqjvN81W5uxicXxsyK5Evd+38gTPjtVK6XpaG
+PridwK.VE.jCv+G+lOE8on+Ge5WeZL+LwzCrK3qrC1tz8dqneiLRvbvlj3c
f+2CAYTxyNJkicn08iYfevi19fkAaq+TpysiTGhXwMLmFIO6ygURDmxTGzUH
iaNdZuyELz8CKQSQhhjGmZCSUqAXZqNq6ptpukiExddkqLccLvtss.6UH4Ek
xX6N9gjk9B8vOqNfelXH1FyPMK9kb7iX35bSBev9Aex75SOzizAzylPY53nG
AZaYZW9pKO70jr0+SYpgxjGl3yN9A+WSEmMmry12yORe6FrJK3lpyVxhsj9k
MYIqlIIGQDpDcgZdo7zSBrYSVJkZPGdhRBDf.Q7KpbBrzQEdQxyUJ0Z7hGkC
owK1IRponbqhkGUEJOPRQNWwyiz9M7PFt5O7rd0M7jAOQqQ207vjc2gvrfzv
.lmX6JCJQbTh6HyPU2SSlyJCxVNwXUYExxsUVJfJy2f3Nu1qJKTb6SwoYK8D
dX1nPLGerpoILIA0j7rZInn2F4L.BBxprDi1HHv2HHSMAwA0gIHFuQOFhpqE
yvNuVTV0PT69s1NxlTJahgFMs6nZwsNSkYZMUACmVBFpIBFpNBF5TBFZRHXR
sUJHXX0DLqNPvFO6BmV5UYWSvv7d6ZBpDPL+eyAt1p2Uf2utmPH8ouXni4BK
1+XF+Znb2UAQ2yt3oW7g4t3Qa9P26Vu7f5GWnDA6.Wnwcrix5ESnzQYZyC57
p0WYZff8hITff5yCZe+hfXvml8iYGVGDCPK9h2yEG4RcccEhY4r2pAfjbGuh
BGHkgc56iih6GZhJNj.30jMrRufe+JYjBl+vSA62ODnTtQxkaTN0928daBdS
VHAaxBIXcVHAO0BI3jXgjzS75YQKpC1GMhjpQx.IFAppWgeJnZdrejFZpjgu
7tqWA1REbxSZYKw4rDxPUJufJTWMzeq2pWpjoT4MxRuU+7Vd5fT9tsl0HVFZ
k1HU7GET34HaGGDlvkXYaPbNQj0YoyE1zkUEKO+hNRxyy3jx6JiYAaihonaX
vpetH8imPd.3XvCPr4.oLfBJ4.rLdiC31hCXTkBjW0bTyC7FKvsEK.dLXAPh
SkWYtO1BGvaqCbiwBfFEV.4reYJOnjEf7FKvEfEXuWjeXWsqAJxdYQhZaJ9C
qFRPfiEGOcnUEHOe6UoWb7VVrUrZtf6wpWUk+9o5RKqxR1GYENIRjbp0webr
A2mDuONoHtgKHWXb2wpDvivMA7Fv2P9wE4MQnxHOpQjG8KJjm+RU2RyBBwo6
kJI437MObksRZS6hqJa2qS1+kmwCL9MM77l1XPMssNMsak5+8YaiUQI0iXIB
rFje5AJjDOntlUkMlrptlQScMKbQmYN.6Np8r1.sxE+6d2RDjFsD1bLZICcF
Sjof8W+ldzY+Mc0no48O3.wWSmqF9peSO93qoFMMu+MX7EqyLExHzRXxUiRp
eSO5TRroFRiIFifjHr17LCtkzQvCR0BLh0TPhcqQo+3bzGM88TnsFcUQFtZB
ufzP7DZLDOAc0BMGiVRm0sQiwRZPcTnEMFhbg5nTe4y7r92RV5LlrFqVp0wD
bLZIhoVJ.LJ3mV1dfHiVS0FBBgMX7swEowwSg3wSk51fZUiB8zTWEjmfQYGr
hUE0U3jmSNsAXs5Imx.mbBCb9oKPymr.mdpBPa4+469+AMNG+QA
-----------end_max5_patcher-----------

The fun bit here is that the early reflections are super ringy but just using the late reflections is quite potent!

What do you want me to clarify?

The process is that spectralshape~ is calculated on both of the decomposed resynthesis buffers of each buffer~ (IR1, IR2, IR3). It then decides crudely which one is the early which is the late by taking the max centroid of the first 2048 samples. It then adds the appropriate buffer to early.sum and late.sum and at a later stage fluid.bufcompose~ copies these buffers into a single multi-channel buffer and multiplies by 0.3 as a way to cheaply average the summed info.

but reversed (that was the error

Do you get anything good out of it? For me it was huge mush, and the single split IR ER is quite ringy, but using LR that way was kind of fun.

Yeah, as I said the result is kind of crappy and it was perhaps more useful for playing with some threading + bufcompose stuff. Your modification provides better results for sure

1 Like

Meaningful descriptors for IRs could be found using:

• āspat5.ir.analysisā does an āEDR analysisā (no idea what this is)
• āspat5.ir.infosā provides onset time, center time, noise level, density andā¦ positions of RTs !!

@a.harkerās objects IRstats are fun to play with

That one is part of the HISSTools Impulse Response Toolbox.

2 Likes