Making sense of (buf)stats

So in playing with fluid.bufstats~, there are a lot of powerful things it can do, and @tremblap suggested some useful things for calculating the “shortness” of a sound after the plenary, but I’m still not sure how to best leverage it for other purposes.

So I thought having a forum post that talked about some approaches and use cases would be useful.

//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////

My first use case is figuring out how long a sample sounds, to keep as a statistic for querying a database.

@a.harker suggested getting a time centroid, by sending irstats~ a center message. This, very usefully, gives the moment in time of the sample where half of the energy is on each side of the sample (I think). This is better than what I was initially thinking of using an RT60 measurement, which I was told was problematic.

Then using fluid.bufstats~, and taking the “mean of the derivative” (more confusing sounding than it is, since it’s just the first value returned by @numderivs 1 ) to see the change over time. @tremblap also suggested taking the standard deviation of the derivative too, though I’m not sure how to make ‘real world’ sense of that one.

So between those three stats, I’ll probably come up with some kind of weighting to get a single number of “long-ness” per sample.

So at this point I have a question about weightings and aggregate statistics. If I want to weigh together three numbers, which are in different units (the “time centroid” is a number in ms (or samples), and I have no idea what units the mean (normalized amplitude I guess for audio?) and standard deviation of the derivative are in), what would be a good way of doing so?

//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////

Now the next use case would be to try to extract meaningful pitch information from a sample. In my specific use case the samples themselves are monophonic, or rather, should only contain one pitch per sample with nothing changing over time (with regards to pitch).

They are, however, metallic sounds, so odd harmonic structures sometimes (example attached below).

Since I’m in no rush for these analyses, I’m using a tiny hop size and a medium sized window size (@fftsettings 1024 32 @algorithm 2). I guess I could probably go bigger for the window size, but there aren’t really low pitched sounds here. Would there be any downsize to using something like @fftsettings 8192 32 for pitched metallic sounds where I’m only interested in pitch?

Now the data I get back looks something like this.

First the pitch value:

And the confidence is this:

Now some samples aren’t quite as consistent as this, but my thinking and questions are more about how to computationally extract the “correct” pitch from this data.

So given that my samples shouldn’t change over time, I probably don’t need the derivatives (right?)

What statistics are meaningful to extract here? It looks like the median of pitch would work in this particular example, but should I weigh that against a confidence metric? So perhaps something like taking all of the points at which the confidence is in the 80th percentile, and then taking a median (or mean?) of the actual pitch from that reduced dataset??

Is something like that possible with fluid.bufstats~?

Here is a more problematic example from the same sample set.

Pitch:

Pitch zoomed all the way out:

Confidence:

(negative confidence?!)

So for a sample like this there is a high confidence bump in the middle, but oddly it does not correspond with a plateau in the pitch information.

The pitch data also jumps around all over the place, and other than that flat bit in the middle, I would be worried that a vanilla median would do the trick here. Percentiles would also be weird too I think.

This poses a trickier example I think.

I’ll also attach this sample as a point of reference.

//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////

So again, not exactly sure what the best way to go about this stuff is with stats, but just presenting a couple specific use cases which it would be good to understand better, and hopefully have others post similar “problems” here, which we could collectively find “solutions” to.

easy.zip (229.9 KB)

problematic.zip (66.8 KB)

On thinking about this further, I guess I could use fluid.bufstats~ to find out what the 80th percentile is, then use @weefuzzy’s abstraction to dump out the entire buffer into list land, and use logic to reduce/regroup everything back into a single list and then either zl median it, or dump it all back into the bufstats~ universe (uzi -> poke?).

So I guess that is possible, but it does involve lots of “in between” steps going out/in of the fluid.-verse to do something like this.

I’m sure there will be other statistical problems that will come up too, so hoping for more robust stats~ to handle cases like this.

That all got a bit involved. Maybe there’s a less baroque way of doing this (e.g. js). This will filter pitches by 80th %ile of the confidence, and then return the median of the filtered pitches, using the magic of jitter.


----------begin_max5_patcher----------
2911.3oc6b08aiiaD+Ym+JHLPAt1KafHojn79PQJJZ6i8P68T2MHP1l1l6JK
IHQuIYOb2e6keIGIaKaZIJkE2su3DQIQx4GGNyv4C8K2LY57rmokSAuG7Avj
I+xMSlnZR1vDy0SltM94EIwkpGa5hrsaoo7o2puGm9LW09+3YdQ7BNfugB1R
WxhSqdjR9KIT0yT0R5tsrzDJW0iPSi4w7EaXoqerftfqmQvY924cKv2el7Ov
.O4ePheAO7ZOksiW0UdlVYKUCW17O8NHJZprwe8laj+baOoxeN9yTEIVxi4k
frUhKXk8mRQHjj1vQ3dPogtjR+ocbvhDZbJcIHmIlvzR.KEDClua0JZQ0ymv
RoKx1kpdIT+wgvPENf8tKXONfwWGN36zU7MEYOAheJ9E059WoEYmY8NufVJ5
kXNKKsFYgChTKnyhjjkhnNXs0Nzwmn5lfvdvkfbI57+DvAPLh6YQlSSDvEeS
rf6IKcEaIMcA8JYVZEDUhAPg9NADgg8fEC5RP7exRWpXth73a.+IVBUJXw0v
m+LEWCB0OdPbjRlLjP5A74MHxjU7ffer6Hm0XPfRZMzG1cL.NqKXPJ8IwaeD
DjKDUyxAyewfAGi.5YC+kbplDl9IF+wsw7BlX3dn2HhlovnEC5GnupU.AdJ.
gLsd+KTwngACNLY5JwFiuPKJEr005iISiyyq07jZuhD79TlpiH2tuIVptI79
lJnegU8996aMtPPkbAItqPCJOGEN80tIaIsHcGSMUzMJVFuopCq.FsLKnmBe
HdJ7wWHAaOxHVwWmjs3yzk03TDqW4zTVZ8sxMt8R5p3cI7GWkkxKYeUM+fxN
+D2eU7BZqubZ7VMw82JXwIUj2z0ErkYoxIQCnV1b0vILOKPqFpNwndhz37S7
xhUYArzxMk1SsqbdbgbkXtlKDUcSdVVRyas+8Rnq3lamyRSO.E4Y4seyB15M
m4cmmIt41y02p6T93tT8ceTrKk+XY7WZh173jDy11lc+ywoLwNPJmoWBPd6u
IMMVPnaJWTjkjzfd024Km3NKELwKnOwVx2nFn5LChGmkWwDMc+p7R1ZZIuYa
730kMa4HoBhl1M2rI8QNcadhfJZ9.MNVQ8cj0Ev0n8yJr+XiNdmzni3zJAdO
sIqjVSrGfUBRnkkJ8BaDKzfuDmriBVUjsUZAq19858dCUD3Z23Xx+LJYCgZS
THF0qMzw1pD01jpFfzBRCCz5YTWIMe4vd7XcMGJdUaS7jJMNFwUCvhy+dGOW
tzjjnOcTkZ4FGanUkxWOhC8TlVKN8iSfbjZADVcNrfC0raOhiGKD+eQ4fOsq
j+JudNsXg3okFRVisWej0duNbULvP8hie+gSjKgyFlQ0DM4xiyH4f0+o9ybj
MTUOx9+qAY0e3CZr0Vyhi0VWU2rpS.enSCeh9Xj.Pggk2ITIFu3y.zYQuZVf
d6gWMPHIFpXDIPK.RbKBRCGQbj9bdA3d0uebJK8CdObWt3GveA7CUWAe.7WA
hKfO7m+3Tv8R7ErJIKliuFz+pwajU3MhzeNWDYj.bgQMkbFeGWJuTXVboQDf
zrtyBkCA.hBz.XkM1jd.f9iHGqV8xuoU5+s11eiZ8nYce6OL5M.L0Ztu+U1Q
PX0UqXExKsxR12zk.+HRcKqvp0.oi86vZ.zkqAJJ3rbpyiSWe0vgmUbj9Z4i
dU+dFv.1FXDTeNTYn3gyMV5R5y07UgSfN8rapS4SPPMXLKvZjwqMjIr2HyqN
1PtqpkSzpnB48OMjUlsqXwdSF0BjEGRoAAINhNmkt2OOeXuvlCevsrk4YrTd
YU7RvxiFhzwWvGpcwaiqfDsS4lodR0UfGrhI3ZoK0zEYAcUA.C2rvFzUXXw.
NKfWyr.NPyhJR7hyBgUsCHV3EoowKyvCG7ow0ruaf3LBrcZ3MjyC0RtcKJsy
bXZrxy2Skd+b4iZO89XLWXLwbg8zkG5J+qxijqSxlGmX723d+UeN2S9pOLu4
04qahzBGLGjzZPUTVJT23o9GRkYZY4WJhJnSEQEeGFho51ipxKf5V32QSK6S
pgf0VNg053fjKfO3SgOPmFBtUrDNsnQBBX6ot6OahNk.H3tvl386w.uEZb8X
jmNxaJXJ.C+dj29dj29dj25XnF9uzzk.gDk4zBYzcRyRemRdGfmAjbm.Abah
v.PHcnHKWvayoIu39HsYxFHb.1Iw8ASzmzTezIn16acKTat0EQmc4HOgwA7s
4fXtJQDApyFNsqXJxjgUQH2DKsptaVMLE0syxSFKLUkLVxfK+jXOF80b9CTJ
Do49vWhMYkkV+jyfbneMHuaAbCFNVPN7Nv+gpDeHYk0lGA9XJ5NveOKUXgAG
HzGKVQDBYhAIrR9sfxrBwuwhkJgHPY39SUu66e2O7wT7cfeVngYsPFjnsVWx
7cpaqvyzL4nfZ.eX230CFKfWg4R436kZ2cQGD+6PAxb01QxNp5OSp140GgGN
MT7ChSO8g5DcmzamdhtJedhFmfY7i.3Y8vtTokqi.ITkfpHyImmot.0MOrOV
6G0gz48.woDqDC9Diuwn.Zk3LQETkxI2qGBoywf.TjS19JrlRKMfTKS06ldn
QyZJsRlZXuPxXUZsK0znWEFqLmw369VRvX6gOx3FqrSjVwNJhYHOcXL5a.y9
8U3xvn6BbhhC+2LX4rIiUxdus5vbu.ZEmVj4DgVn4.MBo.3ExUfR4YA+lIgK
PFNwPsYM8HeK77G+z+YIaKnLdad43m1OZCAwljk.g6SF+AGQ1OMqDKcU14wL
WymYvKes6uCr.tZyJOOuQBtDVfkK8ilZG6QNrYvgrJCiIlbKsGX1rYiHG1Jg
xm4BisVVBtWPsBySE+M9Yweu7tzAPBGVU9fUNZIpGYTZHYz2mdniQbc1gZmB
VcP6Hy5QpMNp7eIIJmL4M5FjH2jRTIuiRePXORFTxXoN8qIBaSIgQJW0M55Q
2iXFEC8fGi.GUq3jbY2KcxYBMcs3r+JP7siiqQAbzQ.brTrVHlKzbwCJMf6M
His0d.VasaTX2ArH++.X3lAsz7W8wvsHu2D0m.3XTlEVhh5fWB0Nuqi6RCFs
BrZ9EOU+97nZvL9PmTrlLksiGMc5XkvvxixYYJChtXFCGn9zp3q+LPT+p25L
gUbZhgKaGCIeKjBpVm4mCZZv5aITbbZpmUrTmdXvAYpgsNwXGxDHGaI7DNj4
ONxRn.OfygpMj1udztXGHx3gZk7Fc33TW7lykKc7yEl75XZiw90DYpt5zaN7
FHJxy5kiAU9gsE1v.OGroPSh7F3IgMLXjKxeAMef67zemGfApfmptbfl6Da4
kHCIqDwVdIxPJuOx1xwhDLjyBaWQh7+CX0XrkVVFuldTV2G4QtKD2Z51acB1
iZOA60IXZf9S2j4St2U8ksJHXpCo3fAkXMUkpNPxGmzLVPrXWVSET5mMUaR8
uWOGS5pShaM8iuH8GZhMBrCq1trnRZTvM1.A2dP0J2mJrwjJidjVhSzkKsD+
HGhEqR1wVJKAIEN7af60hPqULRhlTEJ+wn0oR47wn5svgjZebhpWs6V.djQ.
7ZfalOy.pOgPQdiXQtY9n.fl4c87XDWJbsfpRm3AUDKtVkZ0E0IgCf7k2FQK
5ONLDRGDrL.Xv4qmwgBC76LFfGBwCJP3UwCkEK.2uhFKqDuxKfPCfXAyWpW+
NTUingXWRwhwi+vTXqdyNcDesntNqi.pWqoKo0.xgGSw.KGeTs5mNoUWmVyi
XGA81NR66kKLVdNYr7sXr7INanfWjrZbj3ibT.T63LjNuf8I6up2yNOq.cem
fDdWCRzqwJ.ZAYED75mfitORXaFIWPS91r+H.6fQBYw.4h8Fg1v64BNOK36P
GMNZYmGTr6xg3fhb+fBb+3hau8Ba+vhZWLx+5M+e.l5oMyA
-----------end_max5_patcher-----------
1 Like

Very handy indeed.

I could see something like this (chained stats) being useful in some cases, so wondering if having a more integrated way to do this in the future is on the radar.

Still unsure what the best way to go about extracting pitch from the 2nd “messy” example where confidence and pitch jump all over.

Hmm, I went to test this and play with it further today and it doesn’t seem to work. Or rather, sometimes it works once, but even then, it appears to be random/rare.

It also throws up the following error messages:

buffer~: “outputlast” is not a valid attribute argument
buffer~: “outputfirst” is not a valid attribute argument

Which appear to be coming from the p clip by pitch confidence subpatch.

I only have Max8(.0.5) installed on my machine, so don’t know if that’s the problem, but I remember being able to test this before and it working fine. (then again, maybe I only tested it once at a time, and was lucky each time it worked)

As so often, I broke it whilst tidying up (failing to query the whole length of a feature buffer when it changes). Fixed version at bottom.

These are irritating but meaningless, and are a residue of how jit.buffer~ works. Because it wraps buffer~, any attributes typed into jit.buffer~'s box get forwarded at object creation, but outputfirst and outputlast are specific to jit.buffer~, so buffer~ complains. Dumb but harmless. If you would prefer life with nothing in your console, feel free to replace the box attributes with message calls instead.

Fixed code:


----------begin_max5_patcher----------
2859.3oc6bs0iaiaE94Y9UPXfBrs6DCQpKTNOTLEEs8wtncepIACjsosYhLk
fDclIYwt+1KuIaIacg1wbrVj8EaPpKjmOd34b34h9k6uax7rWHkS.uE7Nvc2
8K2e2cptjcbmo8cS1l7xhzjR0sMYQ11sDFexC5qwIuvU8+OdgWjrfC3aHfsj
kzDV0sv1sMaGOkvUu.OSu4I7Eanr0OUPVv0S.3rfodO.BBlI+CF5I+CI9E7A
yCQWpFqr4e7MPTbsAfxpd+PSmk7ujRT28DYG+582K+4guQp7mS9DQQhk7DdI
HaknAs7roTDBIoM+Xeqnzna.k9S63fEojDFYIHmJn.RIfx.If46VshTTc+oT
FYQ1Nl5gPmMNDEovAeuog6wAe+tvgfawJ9lhrmAIOm7E059WIEYWv5c.VsBG
FY05M5FPm+OAgADTy9E64jTAgy2jH3CxXqnKIrEjqzxtANfQVsrCuAvw+jxV
pVvi83a.+IZJQtY+ZCD9wJIdPL1Jfv6VJwSwW.9wqOFDpjEBCf1fAvYNDCXj
mECxIPPtPPHMGL+KFL3TDnIwVMYzcw+RNQSoS9Hk+z1DdAULI9P2Hhlovni.
FDpa0EffsEPpOfBA3ZXvfC2MYkfE+yjhRZFq1ye2jj77ZceWsGQBdeLS8hvO
ruKJS2k+9tJHelV87A66MoPP1bAMuqPO+dItRMm70jsjTv1QUSEcmhkw6qdg
UHUn1RAOEfg8T.VPTvAnRrhuNMawmHKqwEJVYxILJKufTJX4S3lI29KujrJY
WJ+oUYLdI8qp4GT9xa45qRVP57gYIa0D2eqfljVQdSVWPWlwjShFPsr6pgSX
MTnVWQchQcGrj7VdXAyf.V53hRqU1UNOoPtRLWyPfptHOKKs4k1+bojUbyky
oL1QnHOKu6KVPWuommcdl3ha66cqtR4S6X5q9jXWJ+oxjO2Ds4IoolssMe8u
jvnh8ZDNUuDf71eQBKQPnaJWTjkl1fd0W4ysbkkBl3EjmoK4aTCTclAwsSyq
XhlreUdIcMoj2rOdx5xl8bxFTQW6la1j9DmrMOUPEMugFFsWeGYcAbM5uWg8
mZHvajFBjvpD387lrRRMwd.ZIHkTVpzKrQrPC9bR5NBXUQ1Vo8gZqiq+1anh
vu1EZUMQWBFCQZYgQgZUEpVRaI1uC4HoiASZNTmHerMkF2Uo3vH0wAX7+dGO
Whvoo5iPTocsgs0cpa87ANDTIKo53FgGqh8Xbyejha+KBG7wck7CLd4jhEh6
VZeVMdP8oybEZFBi0ZlsENQ2P3rgMMMQStzdeIe3ZBeUgPUUooiIchL025dh
4MUOrTR09VMvj1NzuwDXOClFbr0NmXwiW73DNEV2MUnsJYwm.nKDAqYf3CG2
xVjzGpXKwvg.xfnwKNRdIu.7n522Ogxdm2GllK9A7W.+PUK3G.+UfnA7C+42
OA7nDEAqRyR398f9H6QeawaDVw45qMWuO.Gg6DvQ2R.WXuQImx2wkROqIFPZ
w0EBkVAfnPM.VYoKdP.LXbBfRNVsxleSqH+Ft82njOd1fxQGwhQq.Ssd7GOv
NBhpZshVHadAFY5tkffXbc6r7UqARuY24Z.7FtFnFqt4Tg8ASySXqskiLPKe
zq52tAivNQCu1QiZ8VY13w8SYKIuTyMBWEnSiGStbyw0fwrPaQlyVU82Jxbv
mCxcUcbXSEoIud6PVY1thE6MNrxzM.rAYJN9Lmx16Cl2cP3DPfGVslcYSCjM
SCnymFd2bzPtsytogmKmGRyQsi2PtavYSCoQZVgFAQNcQwxYgZ5Bc0rH9blE
dtbVXytUo4gtiuH3L2rV6F2RWlmQY7xpnd5K8eEJToXL.FoiMX8VPr14+yT2
opUqzkoyJeiOQ5ezkOo8E7SIbgkKyEl0Wdry9OKeVtNMadRpwij68nceNv7f
WNu+v785DKFNXNHs8vtf5JrKJiUpa+VOAcYldUXfXtD75GDp5lEqhKe8CZzt
EtWn0ssZJmu1HNeM2ID2O9.uEAoaEMkSJ5I.8H6fEaXSzAxG62OL38cbn4hL
NIM1SGaNEtE5C+iXy8Gwl6OhM2EF+i+KgsDHDlLmTHCbDKi8Fk7N.OCH4NAB
31D1CfP5PQVtf2lSR+x0OVb9X8wV0ls.0txquXwcS82TufZdJkC3ayAIbU57
ATFqc4mtWmUBn3Y0PFTeGuGORQFURPICE7yB9cxg7eCTJDu3fnTZ.NXPMfqu
3pAiFo.GbJ3+PTaHkrUZSK.umglB96YLgNaNPngSfqhssIfTZI+APYVg32DA
fKDpHCwNS8ru8M+v6Y9SA+rPl8ZwtZQecB7AW7d4YZVUTXMfOpON1vQJvqvb
ojw8xAu7sw3fonP4YAP0Rxnd2GeKCa92pKICf5buFaoKIQtzijnWmPM7i.3E
5+ao9gAc+MTkUmHygImoZftnfAftoapzAb4s.wwjpjk8LkuwnKXk3nBEDkdh
quJAgQFp8fZ1xVRNziAwwpQFZ480PPgPppLsVJzWikNHWULtSpibq8X3C+6z
3Q0Rl01ojNOsm0rSN22KgjxGMM7LD+G7ZGOJmj9So6cknaxuAuZd.wObH4+9
i23wWJOhzsHoFPFNwHswICmaXAiSPrVJ1rjtETlrMuzordZy47MIj.xeXjCN
dysFMCCksJ6BUOXMdEn8qa3vvk23jQKufjKcRjZG6.9w3aCxpLuEaxeyAwrY
yFuB3VIT9LWXr0xRvih4Evap3+jWD+egR8pxB1AgQeUMvU4yi3AyZyH73EEM
l1djOJbWFXZTvpiFEdvLdCMl4+RSU96wykZEjaRwp3IqzGDMXBWhGopS+Zpv
fUbTrxqYuJHlQwvf7X3QrZzxbIW1iR+MlRXqEmfWAhuJ3WiRlnW.bjpXsP..
jbwMJMfysYE8rZUBQbzP.Vbv20FtYPKM+0vFtE686A0m.3qaoLTgh5X5AQwC
tKMbrVRSy6+T89ClUyObVmtWmmVlj2p2SLbE.LmjTtxcDVkXawtL85hsM85v
gNbVfsMsKwtDKv1thfcZ5IaY531xr8n7cDZ9f93oKaWXnxm9plNbtaSFiVw6
6rIgUIOqSywbOqSta2lq6A1lc2dCxOEn+3I4GnyhV02NHUqFOTVwRcpz44DJ
R5MCKA1fA2ffLtcWQJ5HEpZ3pbY1xotuC4H7sjeHB63IwYsFdi2bbJmzdtbn
SlZVWZHNDdhvVWXHigJxnRvfiXUrcZ.QCVFBZImABq8OHGU056rRMXKorLYM
4jTJGMyyeJNFg1W3117otpyu9S8lW4gJGiDp+DFEVEwvVyq7vvVSqbzUI656
BKBmbEIVScgpCy7oIFSCh02gDaWkR.g7ISQVT+CYisjt5j61P+QlHm.6qJBB
auVJ7cI.znNS5ABP8CAObTcB2pCBM4bnGtinH0HqChe8qqjUo6nKkUdiBG9M
viZYx0pAGQWpxT+TzpsLs1AEsjejIJuQGWu4M.O73A7Zfalh7W847I1yc01k
or7ER06iGyknTWBWKHpb98ZJh0uVAJE1KAG85Kfswm2AWKZQ+4YA2aEqcCpm
u19dg3LLHvBLv+FJdPABGDOTVr.73JRhr.zJuHD5bDKX9XzFzaw7gtcbHBzv
Y7Gl54zaV6wCtMGXeNHf58zzY0Z.43y8Xfk9cB4v9O6Df21wIp9gkF9LtW73
frXbvWgwI.aw.E5eEFoPeaFoqA1EBsYjBO78P3hGIUk6NHeGDdMnJ8XMHuGL
3pLVAmCcYtqicVIT6eRjNmxCv6acUlcCgDA3qBPfsBz8tBikcizoTkV14Q03
sbPNp1tOpttOslt6tdtOtVtEi7ud++Gfk4S81
-----------end_max5_patcher-----------
1 Like

Ok, here’s a pitch tester patch for seeing how accurate the pitch analyzed (in this manner) is.


----------begin_max5_patcher----------
4384.3oc6cs0iiaaE94Y+UPXzBjlLqAuIQo7PwlFzlWRPB5Ffh1cWLP1l1ix
JK4JQuyraP1e6kWjjkr0EZaIOdRy7fGYcw7b93gGdtQpe8E2LYVxi7rIfuF7
FvM27qu3lazmRchax+9MSVG737nfL8sMIl+PxreYxslKI3OJzm9K+L.NEWb5
vE5SJuwWhgEmLYqHhKDebC2zdSxBWEGDMA7t7aXSfX98gwqtKkOWXtGLBN04
V.l4NEdKfBUehwSgkOS710gwxeVMwg2cRSioOKRcxe6EuP8wsVxky1JDIwMv
NH+IkTaZvZtfmdGONXVjlofswpyBhW0Mi56LEK4TFRyhzhOalQQW.FkMNLJB
6TmCuL7YKhsYfMgRR7tHd7Jw8M.CtvIsyKLHSIc536O0iPHLJz7G0+VfGoaY
0lYM3.xZAq2jIRVm84FXKObeiJuELYYTRfnyNSWpuF.bcND.XrSA.vm..rlm
kErheHBHBDY.AOSHoZvbdrTFtIrncMT1v7Te3gLOB5eoTU0R2eXpl8+L3Uhv
07vXs3..0.+6fZk+iByrS..61.Fzm55QezcZuitQmVWOgPmxJ0V6g5lOgiMe
trodU+VYs9GWWLaDBoUV63omD9BHL21PYIQf.T.D7mZTB18z5FKYSGMa557D
OlMJLl2j1ZB0Bs0VYhAgQXHOrKR9oR.F4q4YB8T3b7vw4eYSrMxcHLcjnmGB
AIS81i4eB6om+w4QM1UiHC.Om2qRIOwxyE5eWlx+uMwqmn12BtLeTq+kR6aG
1YjJZxvhSi8JrPFo05RvWpNQQxpUQ7l5mvmlu.gwcOKiWIedQcEf+oEAyabr
G1YReTKAR0F8YFf4bb8JCn08ahB93mKrssQdYPlyHuKxQ2EYbB+RXQeOdq0l
9jt7UCYrT0i1jqJnKkuZySVuV5GxAb1e+QQZvbAPbOGrluHLnQWxwdcvfXhV
xj4iavTbGio3vmNN7mCdOWydFuxRVJ+RXViboaWboKVykM5vwSOW9SaE.4L6
Aw7EFYUdFHLFD.lsc4xcdepLzadx1XQUcEaR4Yxey.QXR7cMdGUQoNvHBzUg
QtL5TRKXDg7DJIbeZxCffGB9nVd3S7zjlkCvcIG30dnWtBjC9ORlBH+sKEBl
wijLs39.o7QR7xvE7347ATb.ir.rbgWkBD+iv3EZQAOn3dveNLhqTOLNvTWS
QPcMvDFccNtoPCpVlB7UiCBUM.2GhPDGs0ORyBdJQn1rIRp5MbCX1GyQnCwm
ZLJqUCj9kPwcqCDogO1ig5zVi9Ih5bRA+pZqImuvfA4fvMSVJGa7AdZlrmrx
ceyjfMapb5ap7HJj6WRz+Pd2VdpvXyofkmJk+gvhm2o7rAoRdVHY3soZrYxi
tElUp9YRVvSi2FVZHttOLmjz8VwRmEx1DL27vpN0hKWcFcsk+TOsebdd5u4.
q.bRwgUQIyeOeQUZdRxFdbXbUo6ZWdAeYv1HwcKShEYgeRSBHkyTMb8k4jXi
WTwDZ5+aRCUlMmeKqRCWjDqHhZcEpSWzbRQDsT.xoJynui3fMM7vRASItzxE
UFNsMaVPppmJ22KbwEEIIQ0uT4yEwWJxu7lv338PQQxl1uXZ3p663YmkHu35
t9s0WI6tswlqdmTnPbWVvGpi1hfnn7wz0+4eLHNTNPjqBgslcgkWz3+48YyS
Shhpwulq7gFtxBoP9b9CgKD2qanpBCxaObSgPzjxd4Egq3Yh5mSDrJq9YxDe
z.5UN01Y4ChuSvWKcWSr2MTKyqUGwVU6Wsy2kVv8BjXRvBsucUtXCAlvhT30
ZpuPnBOf0wGdOccspuqMe3uoXVfbsHmHlTODM0AEYStQ1pRIPfaK.CqSfwBP
QGwMjIlMG5UYqwI3IGUVpRWTqvh64AKtLMdXBJ40CpT2NqCse+kJ62ChKrl3
g6Sx3Uro.DlAhj.q1jr6kJJAeHHZKGrLMYsxcOiitU+0qY6EoYnlNoarzIO3
ftNUBAJkcZi7fWHz7G0xX.oVdie+ElvVyo3VMf8frgzCDg0RXHSB3yQnpVkd
MhPeGW.9ksYhcBSa3opLVqbFphbkI3ImFtg6SzxDHVD8ZC35XhNAvn+BrhKV
pBMcV9IZQSFB18beEOrZd5xu0m9MjwlU4usA+nmyrgjKCvIcwYpzprf4uGfa
Aqncq1uhSR2t+2rExHlILsYNgqBDi+3lTvqze91IgwuA9toaje.9RvWT7Mz6
.+Uf7Kn28Wd6DvqT.FPmvcRa3LlcD3rsHKloEFIX74LcK9x.sRakyDghsBkV
tJChU1p0FnQ61lDafJriApJbRi8L.pTRgF0+eNOYHy2aNkpZ67F8QvESyhbe
lLDt.7zyj1I3gFOvi5wtNAOc60Jf3b99MlK1PMZmfEe1NeCa2WfJWov1ppdX
qI+3E7GGZGGLzPqnjau1mZX67RuwBLnCyrNWLXWXoTlP1R7HzLg55MCNYIaS
mWZAkI9Bf5ryBchpKCR2aJc2du6KIcgIpm1Y+3XQXtsSXnQgvXVRXJkRpabT
HB2qAhPY0I.YEU3NhjA11dD5nRETa6RNTxdc3hMIgwhrhvrSzkCmIVcz7obp
8MjINMHe8cp+1HwWHaGHVzMLdTA1FpfNlTgZZU65jgdiJcbL8InwhJfdGoH+
HRFnqCx.ed5jyOYQp4lnR+xh6LoZ5t.gzn0YROux1OWiGUJQVEkLKHJOgGkI
LqgLl7hcD2vjzWAXFHpw761dAvYRMRukgoN6ttTzgY28zRt6.VIbU8jQWuQZ
eAaDFP1jl61cgoQng4YJvnlJ9FzIsteHCYY.rLLRvSasvhPPKgDqx9eSUfD6
jV7W3++K4+t4wb1S6EDCZR9OA8GI++OR9+ej7+SLYRulGu.HUsLimpx2VbR7
K05BAhDfR5DHg67bHAjJORS1HksE7nOdrIqrd7EapRTYlnaXby.Yh250d1Je
8lnPAPrdCHPnKRYf1Op1h2Cquv83XVfJl5aJGCvvqaLPWRlp7e+fTFluqNcA
YRUFGahG6OhX4PDhVAht1SYKZJ3ex0CmThJFiF.uMFOE7sIwxIjE.47SRDTN
nK.nVHs2BxRRkeFHgVoJAUEDDqe1u9kewaiISA+rTi6J4XR44ZEhos.wN8MR
z2H9gcp.wtW4RgZzUoAqTeUaCB6sl.XT8J8j5XBwA7pYTX2AStub1iolxvhx
tvwR9BktquBfNszyTcsm0pE7Hc8diy8lye+E470XgIYRL5WCjNpTny4gPw84
ZmWJsWOkq0berJo6apb4L45APFIsFJ16qQEHFMvUvJoxjhUjgRMrA0NVnpuY
7cxilpoBstZ.ptxsmis45r2bpCMgP92qo2qu5yinWdZWd8wnKU8VEUFssNqK
uSrhLfUhFfYslbpYE+BVOFYJGFZsfyniWYDfykuLaNVOGJ6rJU5yhv0.8lRS
qHGZ.DnL1GQxq2BL4YRI9XDBBiWlzJ5.Ouhxt.YnZgGly0RE72AvrIkuQEZC
83st7I22+7vlBCCY4004yAvQI0rTNcvLowKKx.uRRa.3T4+CdT9+1JreV+Cw
rYXFQufiK7e264R4blaT3d9a2VISetUlY9jal7iv7OGLBeIkohhzQo.11ZlY
.pDS0PL81QjIwItrmAfymhjF6wb8zQ0oMrAMfXStlZ+mISgksQI47JUjuL6+
ZF3pMjBNfHUsExvUtXTprI4aj2nxfn1VudCRwN6WY0I3496ECg7NW6fxwEiL
C64xT86l5Bf5Y8EvbFzYwxwKSlcPXumAiwz0HRq9sh5uplu8n7e0TFc3yaUw
RtTEqqxXYKK+LZeEYHBmG2CyVumNhp5uLR0Gkms0GIyYDqRKkJHqnBuwrNBY
1V+drQkJrEKXiZkLyrtHhGQpv555bDoAhk.gKajIhiRCy3TL0VREjqghrsPq
7HUVoTKEKnW5EBg8jlNBXcOcDEax8B0T465Ikze6RtlSzwwzNwebuEwugGnX
uJbj9ailjBz5wtioZLjkkAcwTwiFQXijIqWAS0lQYdjEy2ks0aC0puNhztUE
2+u2Ki711DewHDbJ0kAonga2Tyw4z1BfYn12qRw44F6IauX2oocfcxotWGa1
zRYjC4zCKHiK71W8FN+84kW+AqV3lhR7orS6mC.dPVCkR+gNWWGAHiMBTaIF
zIFP85ADrY2FFY1UanNM8F0fPdhWyEKi1FtPspKJdwZXTxVY8WHOkd+YoE3p
oRpsC8IjFPY1frlVXnFAXyZAz+pCf2sOGT.v6vVvqz6VN6dGtzPc5bdfUiua
SxSa5S4a2lTttHRa5sZxooK17V9fhnGxtNvm32j.stqMzTs4LDpfvnldo9Pe
hUA05N+QS0gy..CD+qRXnTQgFG1onn3U8zqVxCTqZortAJDdPTPz3K.JrO7J
QVoi2Q.Cw.lbsFMMks2YuJI0OW8fcaPj8c+IGWNLlO0bEqU+tc1shlO.7sso
PtVzT3J9md5sDyFlpVX82KHGmdS6aQS63eoXRul4Qz40x0h.UOAZXXgWq5XQ
rA.dsQXEQGfFxgXyvhAngv1HY5TqqcfDXrokQih7B0FAFGxPLbzFlbHDLQ1o
cCMHJRQTKZKJavZJTurUs3zcPXCcglxipoWPAddsbsylxsZNzAQYgosrFkFe
076I1OPJLbP1L21PnTzyl9Nmg.N0jaevo6vLp0F4wgQoG0ZyGO+1xJ8d3AYX
lMyIWc605jaIB01V5bM+1wFKZHChEMVYC7PfdtHas19bQOuST1y3f1d6.IpF
Yucdj810QNbGGo8caj82oQzIfxroKrmSg61qN1tHL405cLi69Ad7ViatE6qG
0ghYqTU377jnjC1vUJ7Gdh4pkatIE26aTux68oHjmDVlRPDF1Uej7.Gmp4Jz
7LnhGh5P8gX0sRcwLpi9HOLwgnNBt2ig20VPjuoEf9dPp4H4oPx1p5iI8YOe
GBgUYatPsQOjjVtUoLk3Wd+aEIqRCVDluDyf0Ru1s4RRoxqt+dcwj0R.Mr3Y
pJ4U1Y78amGzUGPvbU3YpApDrixjF4ATGImpOhYNWE9bBOhud+Gk445xz.pG
TZRj4H4uBFW+Yq8PTlChnuUrOwy2bjCB5tWClIaw4J.rdShk2nlHQtLroKzC
BQz5O7rU0dJeHEwzLorAoLlgPw9Fh2ws74TAzQ+j2Ep1VAx3khCDeLx0PqEx
SXlGkTWbXR0swF4P4ebCOF75f3Lvq4qCmkDsXREZr+ACkBJsLdvRYz5iH7JG
G3wfNddliNfW1eDgag3gq5Ol4Hyuz06.he5aD2oeL3QNrvNtcR087n2n20iZ
Vp22Qo7xHsQYPcmlpOirmTuJaVoUdR8p6Q93RgVGC4H6yXjdj1Mh.McT0GqU
gccSV7QGR2lsoomsBzV+XWURzuOKJTNG8+RMAdWxz6KDU6iB6JNSZYcvitk6
0XVOzx.zkJTY4c0sJQuG4C2URYGiH7fwvUVZ+8h7xo6jVZf8qNqpZhCpCZnI
qZotrYB63Gve5CoqL3rP+4gG0wfSlmxFsaa7nCerqRq1pO5vtdwp6GBVHco6
ypXga0YTFDIKS9t9aQa4EKcC6MrDwjS0omypj.8UmBMBDXZxCwGMEtitbJGa
RFIJ7a+XvwSfPrzbUiRRFgfMicvHDhL7D32kx4m.EpQOUPPIFixUSpN7D2OH
0rEKBNZxyiBccMdI3BkDmFJc8Q9NCOM9u4QQIOb7jnz4GlQ6FxSaqFzQ1q2q
clN9Pn2Pw.58uvuS5dxQS9rRyYcjSn6n4DTUMtCBAlZL5QQgVqazkJ658yGx
vxc5SBr9ca87kgg1wOCOh2n+GPFBpuUeB0E5m6ALCMnCDDAa5hY1y8guOPj.
9d0Fe5jSo8MgCqkc+1W7au3+AMRtlgC
-----------end_max5_patcher-----------

Seems to work well with the first example, but not at all for the second example.

It seems to grab onto partials/harmonics for some samples, which is expected and not very “wrong”, but I wonder if there’s a way to try to tease out the fundamental (from the stats).

It also goes a bit funny with sounds that have “bounces” in it.

Attached are the samples that I’m testing with (which are all representative examples of the sample library at large).

samples1.zip (271.7 KB)

samples2.zip (780.6 KB)

I’m just popping out, but could you expand on what not working at all entails? I listened to a sine wave of the estimated median for each sample, and there were a couple of octave errors, but none of the pitches sounded completely off. What does irstats~ tell us here?

Patch with [cycle~] thrown in for aural comparison


----------begin_max5_patcher----------
4741.3oc6cs0iiaaE94Y+UPXzBzlNwf2jnTdnXaKZyKIHEMAnncyhAx1zdTV
YIWI5clIAc+sWdQRVx1hh1ijsSZ1Gl0VRz7b934FO7Ppe5M2MYV1y7hIfu.7
Nvc28Su4t6zWRcg6J+9cSVG877jnB8iMYMunHZEex8l6I3OKzWOzKfFTc0zs
qy1JR3BcaPkW0bIwKa3lNbxDv6Ku0lHw7GiSW8PNetvbWOFcJ7dfWPf9+fp+
hwSg0sIdgtiyl8CeNA0nmiSq5Xb4EKDujn6yISTW3+9l2n9y8uRNFiXjoTLE
xvU2KINkOOaapnYuODfgug+8oncfAw63fANbDAiT9Sx93.r3y9DPRSdm33eQ
7pznD6RAAgJdM.RT+GMzlT.16xy3yeYdB+SiAeyTrJyyWy2dV4a5UP5uPDkK
Rxx1Ln57Hsxtenle8XV4Zx0hqeJNcQ1SCIeSQlQ6.MeaL70Ieit7R4RA1OAP
CuTN0WqP6arwSI13azw0tQiIeuHZdG51PKLkw7bPoUZqBwWAS0qxyx9H+S.A
uPHI7iyd3dFTu2sgWCDvnZiXgVGdO9nK4Pjn5JQeju3AYKk81CQBQd7rsBSn
L2UCHRHJpHdtXaphSk2hRKG3taxxrjjrmVkjMKJQvWuIqwvp5t4qiREyyxUr
SbVZq6pr6wSWX3RMONYcQMTXtu1TQmOw5rEF9QSfSptbVd7pXIrlvSWIdrYq
Ewy+PyefpmrlzQX0CVd2MOFUvs07MwxQqN3t+y1njXwKGg7DwRCihbtrsk+3
f2qukVzbHik.O7FavH3TOo.Hy3R0Z.kX3HpXJETERH+zXvYQJIXarmzoIVxe
LcHhXZ0eOpszvQzV53vdHrWa9xB2wt7dJJ.ZMpGJUbOUeFLHSIa5EFNMfPHL
pz+gUyk9vq.OFsVZUKacwmNamFKSxhD1moiLFeER36UiDLqtPCviHRXINPQQ
kKTvbdpfmOjADVhBzPXMJffVm7S.7xGJQbtFG9D3sJGCwoZAjSNNwj3B2DIv
96.C6lu8PWdsibGr.bdBCDBcJq1xd.xpcgiy4vwjyWdhLa+FAp7kgPlIBFnc
b20fc3UXhfRpBAnx3e9Mngbntlw8zLtu04864e4U4UY357M+6TPLDFggBjXf
IK.DqyFlPuFI6Z7BN0LcQDjLMnDF5b9v9+xIaWkbe4HdO4Afb446Ja6Ky4+m
AUcujuK01s5fGMlV1sluqgjiqBiGoMoSvVCoYDGnEYqVkvOQNKN0taqfZdq+
4nfqmjxln7n0bY.jOvSilY3I33E9B+GOqrZY3MBjpiK0qJqzGm2tBIheSRzK
u9jZ0qCpxgXO8PrISBcpv1Qd4uHyCsaKUVFiQl.sCnMl6k0nMQi4rPmmsdsb
hUGvh+0mE4QyE.wibvZ9h3nzSlQwDsXLKDuaFEkqvHrqAyfq.q9cQefq4Sy7
MyVJ+Rbwoyt9XM61bBT8wt9WA18uuU.jAXDkxWXDi4Ef3TPDX11kK2MAaGVk
UKfAA5q.CeFcJoMXPHc.FWiQ9GyydBD8TzK5w+ejmmcFi6AGjUodG2wWAl8e
K4NfjkpGzmwSjbu3wHo7PV5x3E7z47gY3uBS7gNO7iuFB.+s3zE5g9.n3Qvu
MNgqT+GZzf5aPCLxcz.dMMDpkP.+ggGHHd5.aj92cEHF0jn2UTNR6iwa.ydo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.WWYlnS
ybGyCvt0I+tJOAmCR0NUMsgJIyuQx+R4Rf+YBWN.U5zygL4t4v4GtOVw5Dqv
2.X0R0ZXMdfkOSiRl7Z1OV4eEwp1gdcX.5etJ.8nzpHOd5wrBdi3O.wEfDIb
qiR6QooVvGiR1xAKyyVql+lYJrM+0aEvFoyA.ncD1qLgh9dMxjJk0MFSuh5t
1v3uQKOBjdOLyyuJV2Vy8syHcOcfCqkFQlxLnD2ZFq6AqzxMJt8kbA3G1VH1
I3sgmqVQd0TlZHCZRgxXglkUtJh5JbhuhvoEmsBfwtHXEWrTkW5hxKzsERhM
KjUMVEqP825ypIxD2LBVhoz9raJexaS3TNMqox3Eil+A.9LQvFyT698+lqHI
w3yteGPT+aWbj+7lbva0+86mDm9N36mtQ9GvmA9cUeC8dveDH+B58+9ueB3s
JTDnq1.hEzG6N56JdiYZIWBF2GfiulgGYAvkQ7WHhEaEJqmMLCnhs7LgRm.P
rmA.qlqIqW.jdaBfJIViylOUtjLy2yC1E1FPkmdjeuVSugMlVAoZu4WMHkZ1
cNm.jhthPptuFyYOWJgQMF8fU+savvqSz.dbznwUqhEb+qGmtf+bizyMHPmA
Olb9wXa.ixJYxAj4j8+9ZQlFaZAYfvcjCGMqot+wgrhrs4yqi3qJdL.tEatP
uz70o17c6TLp2EA8MlcdjAxExHXzIC3UmLPNREpXRTn1HQEdtBFvQEMbUBUE
fw3RENOlLVTQEK5r7YiGbc7hMYwohhpE.gnqiQSNTokdFa8MjImXnP8Sp+1X
wWLG4K0zdFMzU+i6jQHzXRF9tNFOlVjY2BDgq5bGhXY4KLKTI5pRXrtIL3Qc
w+llqWoqacxSZYjL6ixxEIpdQFsslR6V3oAeGDJ.y.ImXM.ZVgodqdU8xgqN
FFpVN79VM7qPM.1bdR5xoRO+yiiGDGJPftmezQwHVfo9pZTzQHlcPBcMpjhk
wIBdtkJqB6F13TATznDrX1Kuc2qoxeAV+D9koQOvrwwgl5mff905m3Wqehes
9INyUH6a4oK.RiIy34pkVLMK8y0V8.hLfR5DHg6xEFCHsNjmsQJaK3IuL7qV
KgUdzEYTyMI601p0dUyEoUPcSRr.HVuADIzUkMPOMnyOUQdlc+ioxwJQFrsb
EwtQQFckqpJVfmjx67ckuLnPTeZAMjqicIvgnM.Naq7Jx+FE3PSA+CtVgTIV
YBv.78o3of+RVpzms.H8vIwUoZaDPsqouGTjkK+ajDvkFUTEgQptsewm+699
TxTv2IsYuRpUKuVm.O8r0kCMhpXuF.uuMIVuaTfWi4JKi01AOe0XFUuAdodl
jb.6WO9ZVXEu17aaNO5nTli42FOlo2FeYVFp+..clKlRy8OXmSX.oqHeb47J
C2emueBqrz0slwLqW7W.jSVpxV1SwhGK8ErTNUgbt1Owv6RPFjgVGzHVdj53
+VpdNsggF68MPPoQpp8HixnuAKGgpYpLQslRpqW3i8yzE27HaBhNszAM4r1M
6b++x5aRz6qPmM+SuzKt4nTfbI04Vbbp.FXiLfX1UR1r+StcqUiB0TjtFkmA
tTRzbPu4P0CRuMAwFEg0h30.8wqznJ5YBmiTVcKXR+HG51s5qLBLwoKyFqp6
uBuLmukLu9gK3sof1lbt57rDn0X6IOFuNHqJ7VVYE91KlEFd6Zfaoz4yLYvV
KJ.uURW.3T4+G8r7+OSqdU0IcuvHQuG1qx4QP+arD1sKJVFZ6d4nX7pQ2RGr
kG4xg8Vfo2xxeII578.GSuBJkTltRMz9C76sjbY2ntS+wDY.qL+.cVytHHVo
igdkwX2vtQK1njxdqJeili1PCHdQvuVapFq.3Mpi0bI.v2HePU.biacyG1Xu
xDzaAIGP++5.2JQKi7U+AtE.+4f6S.5xtYWpPQyZ5gv8tsoYd2pa5sY1mUOo
2Rj+9SZ18lJfD2+FMeH11.iSEd6ZcqNlEOqi0Uox4f8pVEYdwxn76TdpypOD
VUecDocWp93.3nVX5PmqE6wsN8cUbBg6s.j8zGolTr4LCtw2FMhm5Z88B6UP
jhM4Zm1f30ey8p87RxQzKcEx57dXPk1fwqd4cjHHiHMPbVykNxTgKBJ9i41W
v4MTxXVc8LmKx+wr75c0dNaToBWURYionYfqXQvnREtt0ZXdiHUTYMzcqEVh
UBWtZelCrbcAJn+xQo8e4t2C55HylPHS87QTe7PdtYyPGbx7Z8ry1y6J7RQv
aXOovMmQuLRMKeX0vzhkuBGL7a37OTtIKNXGoObuNLJgh.Ha2Vp3vYaerjV2
6aSsgCJZsiS5CLv1ACWNItQlS7I06XtZPgXcqUPCt7ayjkIaiWn1MNUuobL1
larubjWRexD0Atcr5tdL1SSLTSbLn5ss5QwQ1sCNt6n0nBG2Agf2pONnZ8lQ
ZngsluThJWG3tD+XWg2SU4bccAOnuiV0Srlhn07s82MyWgWSIcdHgLdVgvnF
untr9Fq4Jrg.67bnYzvCRny3A4JZLQCH6LlT89b6sK4QpswVw4iXmhMjluk2
vgP6uSItdROm26VC2EcJMrzvedf88V64nIo+gZmHbCzr+TpJAnCS1PqI21Uh
eZkZ58RJ1AiJt10p2zgM9U6Hg6giQWicfo8ZkGlc8L900ydNz0g6NvRN6Nxy
kNp4gAx4yRTW5IugnmHtLtMDcDxwQoWaGg7tPbzIH1su.O7hzwvA27hCcLZP
T.7cQZwaPTpCcom7GB4EeW7KLDnGwEyGH+A.8PDW6oWsZc6ipmQU5C4j8Wzf
X.F4hjNhL.8jSAG3COtgiWmEq1YZuKkLzPnj4BbFfGClDQc.eQ3wvvr9HAo+
9FgFDAVWFL0GqMCPeQOE9p7o1e4H7glpRswavnff1WZPny9vDJaPfDmrChfC
hcPG5pgfontvSdCgQPW3HznXCz4IHM3SMyyEmz3gHpMW.2gvtfap.Ch5lKFf
BFEAFmTzCFE2IgtF88qFdcZjLbTfWWjVwCgYTLz4nIaGOtIUT6cbbo5k8NFt
16H35vieqtO5s1+X2RuL7lLisW5u1ctVscQb12pOcod3q4oaM4Xq5LvpMVLa
kZGnLOKI6fSerpL.Nwb2pEdu9YeG.NEGRQp2v4voDDgg80eR9AOulkRioMnp
FQ8ngPr5Qo9XF0S+o.Lwin9Dbulg20WPTnoGfgAPp4SxKgj8UylEktp7zzh0
3HgRcnHkkWerhMkDV+7aEYqxiVDWt4jgspnf6KEkxk2c+5PXxZIfFW0llhd0
CFe014Q1F.hlqRMcKPkfk.jl6ndRNU+Il4ZM3yI7D958aJKv2moAz.nLFJym
j+JXb61NaUqlEBoHltKCB8nLloY3PyOkmec6Z0JJyCQzOANjHan9SdHn+dDZ
gjRmq.91jJV9fZlC4yvlg9.HDQa23lmXaRMwuYCOE7sQoEfukuNdVVxhcIjV
ktacW7Pr5r1ofWK4PBwHeC4UI5gYATBZeToekgZAkNzGbTFssFQPsdP.C5ED
X9zAD39ZD9UhG9p+wLex7Kc6pP72+ShGzMCZSsn4o726zmye0iRGQiwMf3Po
dy.yw9Tyl0tQdJadF0BJCpGqUC0jPqxslyZv9EUg223OsHhDtzGwNJAYdpPo
1pmg0khNLxOuEnctY2TRzenHIV5i9epbfaSld+QuV+oJthWIsrN5Ye0RDdZd
bL.csURV4Pcm5N6Q9vcE29oHYOXLbiCPldQdoaKYjF3vldUUdCndnglrZUQG
GmvNc6.muJcCkyJijG9IKJmr.ULZ2ezOcXytIiZqs1gaihMOecbP5ROl0HB2
ltFFDIKyR7+mS1xq1ZYtGXIhI8YocVTSfgpKgFABLO6ozSlB2QWd05ljQhB+
KuDc5DHDKCP1XjjQHXitCFgPjgm.+xbN+LnPM5oxZJwDbsxo5vSbeszxVpH5
jIu.Jz22DsuOTRbZnzODE5M7z3+hmjj8jcR7v.M8Bgv.aQaFHmjCyX8CEnCE
T1FO5vx.5y52uTN+lSFgY0wk5Icn6ooTTSKtCBAlaB5QQg1yuvoQaGKjaepT
hIrTSiUN6TIdGNn5b63mgGwO5DIfLDT+ngDpOLrbN2LzfJGIh1zmzeiYo7UQ
hLvWoNjvmbN8urM+227+.Gt6lD.
-----------end_max5_patcher-----------

The irstats~ is just determining the time centroid, so it basically plays the sinewave for roughly the duration as the sample is fading out. Not necessary, but a simple touch to make it easier to compare.

With some of the examples I attached I get wildly inaccurate results, particularly the “hard bounce” and “soft bounce” files (more so the hard bounce ones).

That’s why my patches are all messy :wink:

Bumping this thread, as I’m running into this problem again/now.

To resummarize, I have a few different descriptors that I want to query/compare against an envelope follower so that the busier the playing is in my live audio, the shorter the samples that are queried from the database, and similarly, if my playing is more sparse, to have it learn towards longer samples.

So I can do a single->many mapping where the envelope follower is just scaled to each descriptor, but that won’t really give me any weighting, which is useful if one value is spiking but others are not.

How is this generally handled, in statistics? Should I come up with a(n arbitrary) function that combines the three descriptors together into a single value, and then match against that? How would that work in terms of turning that information back into an actionable query for entrymatcher~, particularly if the <-> matcher is euclidian (??).

1 Like

A quick question here: why don’t you just make a label in your samples (around a rounded duration) and query with that as first condition? The label could be devised in snap/short/med/long/drone for 50-150-600-1200 ms (this is guestimate, but I’m sure @weefuzzy has, in his machine listening literature, some approximated lenghts for these from psychoacoustics)

I would use a certain metric on the corpus analysis that checks the overall amplitude variation (to avoid long tails), a sort of dynamic range allowed between the median and the max, or 90 50 10 centile… I’m just riffing here, there is a way to do that cleverly but I don’t know yet… looking at the stats and their first derivative will probably give you some intuition of what is useful. @a.harker’s famous example of @tutschku’s loudness description of a grain was useful to understand how time is an important factor in all this…

1 Like

Yeah totally. The time centroid is actually quite good for telling the overall “length”, and weighed against things like amplitude derivatives would give me a strong idea but I’m still not sure on how to best query for that.

Like, I can query for time centroid first, then derivate, etc…, but having a linear(?) query could probably leave out some better samples if the first criteria isn’t really representative of what’s going on overall (i.e. a high time centroid that is skewed too long because a really long tail, even though the transient is quiet loud/fast).

So I guess it’s more of a many-to-many type question where I want a thing (“perceptual length”) which is measured by oodles of other things, which I want to weigh(?) together in some manner when querying.

TLDR but I skimmed - it seems like you are saying that the key goal is a descriptor of the perceptual length of a sample?

A few points:

  • in general in statistics you don’t combine items of different units, because it doesn’t make sense in a general context.
  • You can either match on a set of descriptors (yes of course you might not always pick optimally, but this is somewhat approximate work)
  • Or you can design a new composite descriptor - this is more complex but has the potential to be more meaningful - the problem is that you have to decide what perceptual length means. For instance, if I have a sample with a sharp attack (lots of energy at the top) but a long resonance is it long or short? Also knowing how your samples are constructed might help - the problem is quite different for sounds with many attacks within them than sounds you know are percussive decays with a small number of associated attacks.
  • I’d suggest that in terms of energy distribution (or loudness or whatever) that the spread of the energy over the sample (which would be in units of time) might be a useful indicator of its length, but you might also want to clip your data in the case that you consider some base level of amplitude to be “active/on” and you don’t want the calculation to be too biased by loud percussive peaks. The time centroid is simply an easy thing for one sound samples that are likely to have early centroid if front loaded/percussive and later if more sustained - spread is probably more accurate to telling you how much of the sample most of the energy is distributed over.
1 Like