Vowel recognition

Hello everyone,

weird question about vowel recognition. I know of course of formants and I think I have a basic understanding of the main current methods to get them, but I do not have any understanding of a more fundamental thing: how the hell do we recognize vowels across genders? Formants vary very much in men, women, children, and yet I have no problem in knowing what an A is across all these cases. The ratio F2/F1 seem to be roughly constant, but that won’t help distinguishing between some of the cardinal vowels.

I was wondering if anyone had an understanding of how formant recognition works on a psychoacoustical ground across genders. (I think that would help me building systems for detecting them, and more)

Daniele

Hello !

I would (and will definitely) use “fluid.spectralshape~”. I feel this is a pretty good solution.
Then thresholds on spread skewness, kurtosis and rolloff. They should be enough if you do not change the voice.
OR, far better if you use several voices, a 7d knnclassifier will probably do the job.

My untested and unpatched 2 cents… I will work on this soon.

An empty and non interesting start here (you may need fof~.mxo here unless you have a gen~ for a fof which I do not have):


----------begin_max5_patcher----------
10608.3oc68s1iiayrledleEZMFfMIu83kWEkdwhE4b4Cmy2N.aVfcQdCFn1
Vc2JiaKejkmKI3je6KuIacgThTVtsc2tSFeQRVp3SUrXwpHq5Oe+6lce92R2
NK3uG7qAu6c+46e26jGRbf2o+96l8bx2VrJYq7xlsH+4mSWWN6N04JS+Vo73
+O+n7ufjkKC98rxxzhfmRKRCJyCdHMcYP4SoAx6R1CY7yIt1+WU2jMEoa42y
jxr70eZU15zE46VKusPf9RdHec45jmSkOq+ohrjU6+wIkKdJa8iepHcQopkP
BCmCtK.GgEughUuglCB9M8OZ8tmyVuJsT1lf5ClsTd2yu+2+XHXVsG71r+P9
fgj4fC2f7ckU2Af3f+Wu+8hWtyQfbc5W4OopGyWRJ127N.ScP4MAcOoA.fQi
mGgwXBAH+Chn6gCJT7JL1CzfFMybytNEvIHUiU2Z4PW1pzujVrkyUqcie2rj
Mapc32U6mHfneOWdihta+gxVqNDb+gJR+RV0umt+nIEbbnjCB6JT.42BIyNb
axWlVrdWl7NoNHmYoIIIpJX.a2jrP8iEbupSWSzJhHgOPr7MXLR7dLBb.N4L
9GWku3yoKqgQuaV9lz0YqqKp23zKSeHY2pxOUWdKdtwS+flDMdRicRd2rGKx
VluVPCMXEhCW8z90.HU1nn0aKxqXcxFC+XtnAGWrbxs713ts2mTH3T2uR9DP
UmrLOeUySsusrJ8gR8o2jsdcKPrLei8SVj83S87auOmexm66dKOy1Osas5re
hKTT9osIeoIEVlrZktyaya+2RVm8bRYZY1yM+EoqS3Mym1tnHe0pFnj5Lewv
YVxEwWj90rkkOIuW0EE3Wd1lJInY64wKydLcaYyiUl731lGYa42UPdsCs6dc
W3OUl97lU71PyKPnxYa41mx+5V8EVIlUu4eXfj5coqqGrww6SeXSsdbFRxhO
ywft+e8q2fhvvX5bbMEg.HjOR.QpBj.Qyiq8WTTXS8hV0M1V+HhUmLL2Ctq1
SVsynNb422n6IN6gU4I7AYG0GB9sparbHIsltQxVVlsnb9WxR+5nP53n4TgF
kn4hCgg5yAX7CRHiApimYEPASXytGoQIhHEHWv0jWjms7G929ie7uGrkqaOo
5yeN8qqS2t8GJDZ54G3y6JJy4ch1e.QG97GdPc4Ov6xIu7k+y7usfOFQo7ii
ByYZoab6w46BxLyfLFLF4YXexyBPiqvJo36Sq3orsLFXBBijVERpd0NJALiR
jYiCJl03IUYOcc8sxGz5koeqlsJSBZ0z38lh0Kx2jwsS+gh7mC9GbMI6xVN+
yqWevdy+5eLiaR+pMARHd.LGEpLyjHQYXjzLIHaD83wm8N7kA2K+uEqRSJFn
cS0RUxVOMxg9fVZ1P6MaReRW2mr9QwPAseWQ7SZeuJiFpSo0mGS04+U3u03R
3ClHs7KaUV426zc4gG1lpASgQOAfVfmxz5kEbCa1a0T8dnKVks3ykOUju6wm
ZetNFh08jew7Iu+Qw7G5bz7hkp48.Fn6fpe.MTNsADks+M+EL.Mr1H846U1w
Bs2MoMscXpZMYuVmxlsos05maZ5aVlBm8owM7T4rMctFSoygo00ZpcX4PAL0
D6.XICiIFEsNKx7j65YBdMfPSSxChlCrcMsmomiy1qmY741r9FXleCN6uAmA
nkYA13226LAGb1fNLiPWlUnOyLrmYGZYFhHPiKv57D6eth8Oewdmyns4MZdt
iVl+nSygzw4QZdtjs0yzdDpNmenA26LVkwgpZ66M9fNeMnjiUAaxy38faeoF
T7CwpdWXjVsRaWQNnt+15+iil08WZ11SGr+z1fBcQzdGbnuAHLbqrMPQOCVz
+.FtMnQeCbzYvCGG.o0fHUlrbWPHQ9FIBzkYaeTDGbUnOil3zHJdLpx.ir39
nKNLBiSix3zHMNLZiai33znNNNxiqi936HPCLJjSiDM7nQCOhzviJM3HS8M5
j8Qn5YTJmGoxiQqrOhkIEelF4x304xHXMGe5KoeaSQvGd.F7w.poqzz7RhkF
8hYQyiXTt8tTaiOM3XTsGmJBLy7cv9XUVGup8Usexq0zbe9.9sKRVkF7Q37.
9+CBnPjinOFDKbOJlnmyAzcvOb.vG9VA7+ctkXAX+.bBDNmI8KsGh63APbza
ED+fdl+ly5YvLlTReudlnoSOC9sBvWvI4zfGSKWkbe5JGQdpdJGQRWNE4gJF
T+3NKdXbG4Dte24F64l2uM4wzdA+JXO3CPWk4wJu7or.NjMcHO6shDuJJDRb
eaZ4eEbHdDeetHTDyelOmyLgCt8hiPzLl3ISIDKZXVB1EVRk652GR2Kddzlf
E4qxK3bBG4BD0ZlAij5hBmtQBXS3Pv1ZJc8RhYXbPukLjGSrba6yyIC38jg8
fh6dQYHOoXzaJd3QkVdUgpMPVNYkPJ1tiUFz4Jd3fEebxhyNZwSms3fCW7yo
KN53Emc9hyNfwQmv3tiXb1YLd3PFebJyXbLiCNmwYGz3lSZbyQMt4rFmbXyP
NsoeG2Lfya7xANd5Dm9cjiM8w1FY05065HrllzU15xePLwqeJ.Qo+nsejwYA
HGkEDpV0pf9Fy0owc6rXXvyrem5e7WmsM0nsOls+4Lys9iUAaWw6KEf8lGAY
RiS62yDNMkgNrHfarHjyrn6txYSpITWjtLHufa2cZvlcEaVkFb+pco9y3.p3
8.i7iyQGlygiciyQ8hyU+eWjbQ6SKuIaDLmhn7gqXTF+03n.3bO3dDDVtFUU
yMDJssaZ63gY80DjydpJ3g7FBI9NYaILJDemr8.vwHNk0OE85QGq6r8HpjsC
IBXh.8iuihUSpnhuGO878Pm46Qb1N.IY7bBBEdmpEAYg2X7FX7wgBFeXrfwi
YX+X7PFoQG9S.iG6LiONjv.R1cnfeGcmrEozBbiw2sGOQnhm.jLdnmZ50iSe
536H2467FBfJ0zSDp2kc8w7t6D1aH9d6EuuaLP05pAu+U2XefgYeV7y0Qg5V
tTKq5+lTVmc.vEDqS0V8eFnQQig44vDPCotw7.uzbDydfSrKm8zOCRPU767i
ssMeWwhJIxJw7.6.+xzskYq26Tze8vTPD+nShn3T0xDNgv6VlPv4xukAFQKq
BNtnaYRAKruRirqkVFx2VV30RKC5aK6pQZz29YnqhVFaLZP.WCsrvWssL7q0
VF5EpkY3hMEV5Yh30s7Sp3c9ojxxhr62UpLPykX8OIwD6wU42mrpUPpFJ7Yu
eXf3BcQrI2BMNuH1jSd.GJW1lQQS1Rohxd0rH1ri802mSpUT086d.sJaaoCK
9ppKM3m4e5gzhtqAqsOmrZ0H3lgn4hzRj38oZE.QHuJVYbNtdDkXdvGfSFKA
p1z9HHZx5fQdyrrn28GYhX.F.bDxYnZQMjNcqNTLZhWJhUuKjltNVvnKSJSL
udQEadSwY8qagNAxAwSlhJzqykKpMG5ZW92Em4Nnibar8Xmpg.jPqQxY.WE1
iaBeAYEVcPae6KFtftv.KeXGCHnGSGlc.N0Xb2cdpU2u1E96aJOFXBclpSrU
2sZyQIHySyYTBOGK0Gg8j5E6CJgugtLndjmTeL8RB6g9h81YVGRTMWoMMvvM
M3kQSCLldLWJxbrXOod1EE0GYyk61n93KIpm4K1GcIQ8nwH2itPndpuXOgbA
g8DeswnR+6kA06qjykjJGLxWUNrqkAo8to0Ci74rk5bljZM3AiUqAOp3M5gu
bMLBuuvRO5VZAKHpbsHiPz4X1dfQ+UmD3acQsC1gqA53OmxrzhaA1vVPMZ1H
a7slL6WjDD1fYGrhjr07IW6dFBCQ0KGU0dyAc7YHrnaYHraYHLc21aYHLkZu
aYHraYHr1hDuoyPXHWiSUrxYvLpLZsQSWvZYuYxQU9mhLT6QAQhAiNsoHCvs
TjwagTjQeVVLn0EdXggOVY3rkFdZsgCVb3mUGNZ4gyVe3rEHNZEh6Vh3r0Hd
XQhOVkLFKSbv5DmsPwMqTbyRE2rVwIKVFxpk9sbY.qW7xBFOshoeKYroO9VJ
x3VJxnC25VJx3VJx3VJx3kamSK1szf4gxLlPLK73xQFwmybjQ01jVzZpxZBw
hCAusk4Mv3Y.jXWlCCE657nvq3jjAC.iCgx8LOTjdTtS0hPxT47MFeaFeDCt
mwCBupSRFTTLQx3o6SWF.Ydw4Fi2TO9HQ5w.RDp5AHx0aRxf2PvPxcpFCmgK
46HLWZ3VRxX.F3sjjwE.q6VRxvUNxsjjwKVK6VRx3VRx3VRx3EpkcKIYbKIY
bYzxtkjLtkjLtkjLrBDuRpnbP.UUQ4zKf3orlxExtUS4rC4DPzbHZhKpbgQu
EKpbNuzzfLrppxoWaZSWwqLL9VQkqu3AoWeZHzDWT4BQ2Jpb8h7XHQstkBm3
hJGk9lSh2mDjipN4p2eNSYBnB+lNAT8qve67jCpz08X.cxyAUza4fpiJGTo8
A9DlCpnuoxAUDmy.UPTT8UkCY5.b3a9TPUO0rT0obc7djp6.iIyLOH.Y5TTA
dyVyRGJMgk9sjm2rJ0UlTjZOaBTUUbDZBSngw2RTXZ0TSdlBKJ7VlBy6LEVk
2NhAJucLgoJrH1a8TEVTnmIBC5kTZ7f4aNcJ7hJ80.FC0eoj9ZBiGC0eojl1
pDDbO80vtfjbB8NsMAuVxkYd2z5QN7BK41D5aRGhAtfj4nTeGo3hh58McUEd
I0em5aJwjdIkpvn9J2StjRQdRCd7J2KQuVTH4eSCcsLLhT4iOMsvKoL7FI1W
FC1Yp+V55x5TW6KccUMkOaIiqhzkNj4sHHUTLoQ0c7L3Hx7VXP+YdKvYKUls
GZTQP4u7ve8FcKIShYgXU9wGbz4rLb+.GZnTV1d+ZocFy4GnKCtW9eKVklT3
.jFKinKV4JwniGQg8injgPz19du5cU64EAgM4kPSPWnbqpA62+f85avFord1
jm975wEUVbO04Qlk2IZohOOrxSctoDJEaoQGs7Zb3QB6ub87MGazZ46oDtVz
7ujHVWGAe.E7AbvGHAefF7gvfOvb.bg5RMBVUiXXNlSHQ1QWD83EpeAP1eOq
b9lux6Rj+0NvTsTQTcnRtBvTKNCz92Lm.KMpvUE5eH4vaiVDtyBWi252OIi4
viaDNNz7omS3V.9Mc3a5AetPTmH3lqVrLfDr3ojBwxX0vRYsGVChLvlP08AC
ilMwn+EfZl6Kd7dNrDW8OWzZykukq2EUVtAb7JVhNsJVZPWliESSH11r8ZAz
cKSKF8AssMyhgsCmyBAilBYdPgvyAAF5AAFcNfvnKdlbD0Glr4lSqjJNQMtH
Vmx006P.w2dgaanwz1dQoPnWxGfyAEZZSEZqXFEclHPjq5.NWD3ksPno8563
IvZm7vY1+oCMf8jUKqTNPtCX6W6UXM2fzkYKpO+ciYXHhbpiT0x6QmESiaZP
hUiQZXhGn9Bu0t8cCtzoOsXylfmS9bZfnPT1sFAXzVX0TqYRjQk9k6Gef1vG
nc7A1K9z3YYHG+VShtub5qsb4a8NDVxcuVxYu1yUuCmidskada3CdGxEudmC
dsl6ccHm65Rt1cvbrqi4V2dxottkKcGHG5NXtycvbl6.4J2gyQtClabcHm35
Rtv0mbfaO491Ay4s8mqa6OG21etss2PJYKbRlygsVxcsNkyZcLW0ZNG0dJ8z
AWStKNplTyQUwridx2rqCu5MXjRVEbuynmNZSthd83THF4HcJzkP7lru2prh
25cn.+RP7q0cwV0VQfMA.O90gXqJ.e.mCwWirfKbJ.xABxGcrA4SFE3WzX8M
.XuHe25xTW.Ysm70RqgP2.YZOfL3HijpFLq9W2Mwz4AR2DrNOa62cuRko0+R
kPZzwGwuH7s5T1kacJKRwsQx4zfUyBEyP2JUY2JUY2JUY2JUY1FPxxppzgku
WyQm1VtLfC3ttszHxnZH1pfhcmFy1J4avwm5L3uCa.Svkd0GivsPc2Vt4Sa8
bq3KQSFX5J+XSUJjpV33uUHxtrJDYMrafFqr+VrBltUHx70lBGsqvYaKb19B
Gswvc6Lb1VCOr2vGaNFicGNX6gy1e3lMHtYGha1h3j8HCYSR+1kLfsIdYehm
1nzucJ1zGer4yc61szbDWT.ugtNMPMBUP156B95SYKdJ3e++9yAkEobsXqeL
HYav+26B9+Y6tIhbrzqG8UWELsz7UKpKLs1lZ.imx5Xli4P9KvDxuqLP93M6
JCx1FHSDmKCdLYGWzLYs1WIAbkBAIkAffmSSVOO3eQbwqKyCdHeWwcAIqWFr
8orGJCRS3L8MIEkYhtfA7qHIXYlXCXvoB4O1ItOwcterds9HsFlOTbn7uX8e
2EDNkkS.z0qjfik+LwJX8wU6RCPAxcXBFE7yKxWs640a4VU+yEbMUAP+qtJp
J5KNF6W0Ug3PEdXpqfc8tvWu9Xl+sfeVrt7+HXtn53P8pFnoLrUul7AnIux3
Piuw6bk28Q+XdX.XNCyX.JP+GsFqLb5YkQ2XkNvJGAmTTXxZOn1orOI6FizQ
Foe7wp5LlZoPDM8LtvaLt9XbaWjscadw1IwdFUDsoPVytlhh03jVBkovItJu
1jo592tUp3bnTwwnSboha.u9p7Y4a0hFGfbhJZbrvaEMNiDs4T3huErE50Pw
rhBGQwrhxtVZY9V.xnQWKsLeKZbzqgRXHcLEypvqgR9DkM1VF7RukEM1VF5R
ukEO1VF9BukENlBpI6ZX7LFcLiTCuUZ0ZGRvqtRq1l.z+5g34r0q77dy0uMF
OcqKFxs0ES8S+ZbcwPvpbtATuKJjeiPA2VXLMtjaKLlaKLlaKLl15ytvWXLd
32YtFyOGf7IfOD05YIDHybQQxj82zFpfaQJnONlXWBUeAPvs+wqXDn2O7PDr
VDelzvBPtw.6iAJKsp+r70s+mEk+vGQ+zp7G+gr0+J32lug+xO9i+z1r0+.5
m9O92+opiB+sezKdrJGsEK6ch6upCOpNo3a73ikGuHe6wviI5RxWEOlL47X3
aPdrqKIw+OaEqYzxmRC3OuO97tzUqRKBJKRVu8g7hmCdnH+4f7Gdf+O9UkTF
70rOmsIcYVhXYG9XZYmI8p9I6VmI+84qMNM38gTYLKE0Xfr5ZhURJpU61jtT
TuhW.hiKVs55t7IHVsnINVsVKJZusiRKCLmdZhRK9VTZcy2uiJHsWAQXAOx1
0ktu5Iinc0OO1dke4pugicqgCurin1XB6D9ZHrSnQzxH2h5zUeTm9iUAQvXT
PVI2ncjWgaBwjasc6YKHmlpUi8HhCQaB4T8P+ZnVOKlc7t0Vc6XOw4KRB7nX
5jEmuPzDUNm8bsod4wQ9op0MN1OdhJAOvBmtZUOch5LbkwDpzHsRDhOuzGQv
xMWHc5THgieqoPZ6lrUqB9YQN3aU55GKeJf.hCcMYmny8VPEi.SlL8S3vaLB
OXD5NDm.9.8UCevletNt8ivf92pQdlCNQqtFq9zZZqA8mJVgUeV0i+FArg8W
kWh1QNXBD3TivcydaV8G0zWOQw9V6lwWRUfWb3Xn9KkJMON12JOM4Ztpeytj
pa1dWw0wzKMpG5C0GdIQ89Vw0itjp.yLeod5kjjCi4qNmKIrOB5qNGxspeso
u8xWVbcybZ2qKtQSSAZ8Rs13Zz7XS3ppT7fUYXKWA19JNkfgJa3SMj8hUUEY
.Op1VLCQY8jWssX9TtsXF1OZuLTnq07sH7Ynhkwv9xkewoPhGXHCbtnPmwvy
BW1qx2H3b.hQdIHRNG0HUepvjLneU2uZUoMGLc5OO1p6zvlJYxLoIsHD1sDI
ev8tIKUkAUYEXG5TU1CCkKbYU54aTUgvFY3ZOKxdSHtvY1k4qcpMCiTAxgT8
p2UVvvP2ZyaRJ3BIkoEeRsUdZYZiUCFuDkWH55hgJ2wcDBLMRXimOAlgqXko
eqydk0X87LjVqdUxhGc8pjBNtx44sRV4sRV4sRV4sRV4sRVokne6VIqjnpmC
DRzbVDB.ipR6lhb6Wrs8hoyUvxNoowW0UvRMXhiUqeIGgudbcDk9JnDVVKsq
6h62HwDYQ9AAlSPzPHoRhDKqJcGsDI4HkHuTpue+wpfh7xfN6pt9DLUawOWq
se8gh3IPv7RAF2thOJSmUOZe3HNdNlxMmBWIaFcWPb7wipnWKnp20f1JjMLb
NChh.rJjkNIkjVH6pXbHy6VzZCDkGrJ49TtHqSQx.pJOSD.bdXLCEFpSN5Pt
pzPqEQMmA0AFaGbY.XKSJSbDunHrZBRgFvK5QiWzyOd4bPvprKzDTLAwDCeB
hHF5RvPwLw+4RWS85qF6UYLtGDkLfFNrikw3ZucILPhiSiQCmXkm5HfidnXB
80vDW3F2rJO+y61DTjtLHufOSfzfM6J1rJM39U6RcAYYJWmq1Z5vnI.Ze0X6
nnDwCbR5ThgQpDrAL7n6riiNRHrdgf+Jq9uq5oCUlaOA0+cL60Y8eWn4L.6j
UhwU4t.YwqxQHEaGRQQutTdZoN1XL9H5ZyVr1w6G+TsQguVTWt6OxBP9nvTT
1mEkOFzQ60GzwZcTkByKMEmiRHESf0zgNIBoz2tBo5wyAviWJEbsJk55ZVUi
YS1hVkbaQq1bBPwml4qShestDVgLOVwVzywhaDALUBSrPgnyFE5JFhXmCJzm
E2nv70W7kALJzWJr65c84L0RbR2UAoFlGQ1ujcn5u8R21X9HAGdNjO7QK.9r
HA6yJKkZdgx1V9.DUkNWEAbMFu+auvsMrOnO4bf9XePeD3bQgPejfQCIefCk
QhW6APwBIs5auvsMhOq6ax4XalP7Y7ER7YP9f3kN3nyEE5pDLAetnPujCewo
PezSgOGbYJxmsYB6bQgtJGRImAsMTezGRQmCLzms6jrd+8higTe4xfyAE5EF
9hSg9LmErCa1oNqI2Alz+9kJ9yIeKrSMZpZUk2eqR34uE4qxK9T1ZwdAP2.k
9L4vKPCNO49Gk+NsCXzWk9Eg2Xae8IKVjttr1uALORbkQ3XDLT7IFHNBPzOs
lH0cs3XE7aksZPyrm4s5r1qoj9j.Z.jFblmaPYilFAgonXQqByPTh7SHFgBO
2MutaRnWDYkGxVsZO.8NqOC9CQ6cxYOVjrLyPwCPgyv8xPhRxN6Niex.on+4
npeNKJhyoty3mr+yq90nX.LVJ6h0xthOAk0Fdi+5j0Op21DrtEuoYaJx2jWr
eelLGG242uqLeOvzIGO2NAMzuXkYsFGmzkIO1NhdORYIBuaC.I+DGa4Xwopm
ya7sYZO6NN0BYUrFB2lV9WAed8Z4OK6guO+ozUalyQiRwYca+mpmIcHZzaaN
XOaFS7fa1x6DMn7jxVQa67BgNtyCq1RlUX3325gP5UJFJW6ulAQ0obayaVAg
T5ngvnqTD7.tkkV7WA+LmzWml83S2yMsca.I3m+p3akMh+qwUJrDAo.cneGu
r3X6NKVZ8mPfruMR8CYk16DaSx7tfMEoKyVTZWEZ2y7bxlMYBAl22HRflpUQ
1YRD0dfEgizaJK+2D2zKh8v8.atc0FC3CvfOfbTMvQu41IgSJvr+42MduGvJ
SyUsFhMbU1vwvR1KCyWhna3krMIYvolJftREvSIUfbkJPmRp.6JUfOkTAwUp
fbJoBpqTA8TREgtREgmHp.CbUeQ7IrmJD3pZKH3Tp2B5HUPOwz.zIj.dhoBW
TZQAmXhvEcVcBBT6H6FpVkxXhLjtHFY+2Z7q5VOoNKMGXnysGYIcApyp4huY
t8fNusGpesGBNZf1C3DoEx09+rSoRHRniTQzIjHnNRCgmRZv0wlHmbpvEExg
mRUggNpOd+HomFpvUNB7TNBYXnyFKbJGhj45jbf1EOOLEvtg+6.kYOreF7z9
e99iK7M9DluwFhuZNcuGO32068lXa8EJu9gmQE9tSFo2wcGmFdqwvxUOnF8F
NtiHTbGQX3FeH3rG9s9C8VOgcqdjXrKJzsmp+RDsCslGRytDNswKIquGU2fg
BallHsFtLSgJaV8zqW7dN2.QOqcjyT7J4qJzVRJ07+nhz5KGabvmiEIqWl+b
vWxy1Wh4L5d1Y0Sefex3UXMs+YLPP58ckN+2q1T0rCB5F86twTkfYP0TZ4vS
HqQ9Y0Xa.ip0F5lbVGtMT45UytcsujwZuIhUOanMB3xAQi+V.b1nXsJXAgXG
RfZ08LsQuR2.VH9vZg1.jC6atoAOzcU1uKIGEp.ijAxf1I3wCiJ3yKprcCus
TjucQ9lz+Zuph7GenH8+roPqg8SsJ0kp2CaLkoW02kkC15wvY90NGWijO9nX
nU6MEE6iQBGYWd7T1kepkvWlr3uFkvMTsIkhPgirKODdp0luJ6KoyeLIa89V
3WRNLVa2yJxEnsxAn8zx0h1D0fZ3vgjHV7Tx50hX8aQJgXVJgziTR0sQjAX0
298CQCp2n1878ZGNYStZa1iqE7a0N5dezokgqde6RYezdCi5wDoujrZWZ9C6
svqx9tZMir0YkBQL4y+irZ1618h51labUqxW+nc9ZiK8YtDh3xBOXDciylIM
o6i0Mxt144fYQoiOoJv0z8YGuYs2HUx6qaO9nL0SvH1Uk9p3lgtcypjuKxJ0
0VOBFKYNH7bL+OBoZaGym6AUmmIYjNCXIdNp0RvCka4xO7611YtpFDMIClMo
Z+9iUAh9MO1m5eiXTXnZu2w7dL8HpOv.xJLb2DCEp0YiZD+jUaeJgOje8htq
6PiZ8dAgv9wFShH3IAapViMiCerMsItptGeLsHHecZvC4O726cpSdMTpxpYU
wSeDSKZe1m3jMPJe3.9MouVPUUGmoV4Ndy4Yigw2mgSOjW7bhj5B6qKz3mAk
Mwjko2y0RF7COuc9ONJCqv.RMrTm0r7Rd.cpkGr01a3RA+5BDo1q6pUXqxxJ
uZzgmbqIsn27m9Kgq7NFSn0i2h8e9gP1jLjp11uQ0MXGWPX2vs9jhh7uNfc0
nHYVbEEoF3..5fGYkoOqCsvr+24a3yNOOQzK9NwK5Cj19.YsOPd6Cra+A9mV
UV6NJ9VZiuk03a4M91g6x+BGFJRJSWmWjX5folNXloCla5fGdP+RyGwuz7l+
KMus+Rya3uz7V8OyYnIM9VZiuk03a4M91tCyPHY0p7utrH4w575AUX4kyevi
cZy2cLVM0LpFG59CFUOeF9vH9vXu62eNsj9yaWksbeFeb1WEkFmOm9c6gZqV
nL5QIXEVHeCS7.L1aO8SMICvbpNXJl9Dr4SvSqK5KI7NkFkWF7PvCGyXKP0p
mlqRUlx62m26QdZeVzjXX9d2Jn9vThTOWleb.kNavE5soqwmwNiVmqxS74nj
vevaC9GyRRyx28OlMa7SMAhUUgBkCu7wrLnQUU3vSucYa2rJqL.DfGmFZD.a
nKiJ46Ozr9wFvg3IoCzwplwl7RdQvVd6OHIXcdY5eebcj.0TjGF5+zVNWVv+
e6+gHddf4iSNAaP0JRY6J1+Q1gQSh1joW+Jef0UAxIUut7SBOKtZ6mxV+ok2
O1HFpjPz5TfHh25codY0HoWGocWqxSnr49I49CskisO3oG59n9qurtq2ItAI
Rx8vbU1eJoCx0wWBbXctnRS6Pc7k+nDihUwZNVcYeTOOQjVDSeT0Wg5XRDq+
8p2fDPs0KgokqQKBM0LghoGHzPEQAZ93UeCoaQeTWM+zjZn9npjHZ0x.XOop
2RddRpYlIUcRwtNoBIZhBUGT0Oc839fVfZXM5eOkRH6OpOTZtYJkTCTYZY.M
gBOHQvmarljP0IzpKUeTybeL0OBcmEtOhtmPoZxUSng0f4pChU7dVSBUy6iZ
Rn6ElbjP0SS2yNSjZ78nlxsZITZEsGZBNC8ryj18AFY5f1hmvpdGJMgQMAyl
jItIDSOdvLa3N7ZgIlwN76Y5MnSVS3bBnSu6CEanKDrQe8JdNJxDOWCwN2ER
6HHO6+fp0+gha.krlTok9OD+fxlNnxDwFF1saDp6XRrJ0m357cPCQYMjhwUP
ZjeciZ53LireRa1ObuNTS8lfQ0E..81Gq1Q8kZcXnoJ1IsNf0Z7dMJBZghgM
TbE24n9Rs15ZgaSsQZ4.HvjhpPiPKxn5KxnIVacw5.sg35rwHiRssH1l5vHi
jX+k95bQ6z4BD0syk9x9HqF8rWpnt1JtlkF3LwKhzggnXM082nSUXCkXUfIp
wktWLMZbcp9k95NE2s6Dqt4msFwuYm+8HWS6.HcDEbmNyGFOIMsyCLDbdf3Q
F0yFNBxb2vC4q6B0PSXKq7fr5jYrQtNdjjoNjDF6A009I.oVuEltZWQqYE.D
pWytMNHpkJgCSixYZL0UFNLDUmMp43j5lpzOCG1TF1KjzV2mNyoCVgEJo+3V
xkMUvSMpfmzvjJuHSm68TQkMlkTEVhpGBQTnoQ32KrR.9SkirySKpD2nOdHi
1F0pMAUyirOxMvP5y2qpr7UKlNwR8eQ1V8FZHb7KDMaKtn7G9q.p.MBf7WEa
QFd6l+4Q5bV0nwTjJyVOzJMhYvOagm8Hm+BAUJcEj3wCUj2HPUUU2Mb7PE8s
BTEpJ3t3wCU32JPEQWKmYyiZuTIGEvcoha6VuIYwm4vk6++3VebHcsbWWcCh
7NxAH+BxB0wvYeV9vINFOiL5NU7HkI2wL+YQzWWA2A10ezx0hBetpsc3C+ED
EBi6DiGvbXGGTxeAf4yqXxCxCraPdTO+lSPgeLFKBE0IJO7Wfc74qfZgzH5j
GmGX237.lKPwn31g5g+h.2i6DsmZm.0gpCwvIOjOvNtqlSyQvHbTaWVyegES
v3NA9o1IrJWLsg+A1w80bQ.ZDQ.zM8MfPBfKaf6DDHsnQTWQioMFPF6yEhAw
TVauBoDsaFEHAYxaXwFwV3zFHHX2.AweQPqgscwZCRE2ApoGOplMrhglACpg
hgCbeg7GtSzfpAqS.s5eOLLkDIJths6gUID2JtPh1QbDznl2JLeBiNjotWU5
wZFfnZ.LqibqwtWJkXmfnDA6FknNCtw1SGLtJKZGmXdPbuU7hDMGDUxYl9PF
A6FxHtjPbLEYpemfLvcBXf4NjvSRPifcCZzAfqiYDU.JnCfp6.1L5QsNwzF.
IXm.Hw6JxkeQQcA51CPC5LXWW0ebdF9jDLIXmfIIYtQ1Dq6PysTCRFOM6PLk
fcioTmth6C+DgDFFQ53QTQJjIBBXchuTMFvTFhIX2PL0oSXXakgsBzzgePqX
M4YOQGhzDrajlNXEbGyJZoyXORB6DsoVhJSU.mfcB3TmNd1.2CMi8mniR65l
LOUAeB1w+4R7ps4kPBGcIch9TWQA73o1giAEraLnNzypYbnNX7QqXQU6DnV5
UfSaznfciF0AtayvSTinFRb.1QbeBCMEranoNzUItkDbaq3.TaCcPZad2DFl
JXmvT0YlcGFFHJliZcBVUcMD0FRGNsQpxROsNzXkCIZFppVfWi4XawRhWr.V
Ic1Sy7QuxyXsyNhZ+i0IqHF0HygZKqA2Hi81Navxvnt97MjKxJY2H7bZzgSJ
EB58jXFRs.6iaro6ObNK9Bz0VLLxgVbibE4gDYK9ndxDhCOYTsDB53eRgN7j
pWiRG+SB6xSZJZSPlCOILrOIUwTwEhe5MAEAu+acY0vihXaTsiskmQim.TA4
.nDMAOm3WnmiKcNgl6bRN8OX7oPsfSOYp4mL5z+jIlexvS+SNz7SF7BnFA6t
ZDyC+YVoxQR5tnUudMId7OIxK1Sh9h8jvuXOoWpGD1IAB3T7jHt9jfSsTu6O
5IWKka5JHmBUyXWjVinSAqE4gBwSuk2Rkbz5AwtVZRlv.TYp9NFJR620+Db5
02hjVwgbgfglI3HUFIW3JAPrLSqyfrPBqVpc9HoNhKTG9bQcXWnNzKO0Ac1x
moRRDd7nYrKz6YSPD5B0c1DDchWewJHRmPAQzwilgtPumKAwHWHtykbHyEh6
hULDOghgjiGLwtPumKwPpKD24RLj3BwcwJFNgRgziGKcgbOWBgHGnsy1PxNP
amAQPWbGwj32VWdPnovipR2LOHZOEOIDwgmDp6bVUAwJYyFQgQRe0xGxrmS9
cEaO5N4WyVq9pL9XyJR+x9XiIWIyyRJVHRTeKJ2UnVA2eKTssdlIVm2Eq2ko
sEWFYtVEwKUi0bw6pUs74Oe+3JpStVrtFSg5RGrQK0Fnl0En5bWaEjKyMYuK
BWSBY0HsTNc7hNEWqp.wZsnZMxBp0HKlViqPZYtHZYu.ZYo3YUEmYyrul8Nb
iKVu.Y4fz0PEEK+jrTJYjaNDAQscShR2kbOj79+q2++GPP6M2U
-----------end_max5_patcher-----------

Hi @danieleghisi,

Now that the MLP tools are out, you could also try the FluidMLPClassifier (that’s the SuperCollider class, sorry I don’t know the Max syntax). I’d be curious to hear which performs better the KNN Classifier or the MLP Classifier!

Thanks so much to both of you!
@pasquetje: I don’t think I ever had the freqs, amps and qs for all 5 formants of singing voice, your coll is very helpful.

On the other hand, I am still wondering on a more theoretical level how our perception works: why do I hear an A both at the bass and at the soprano, although the formant configuration is evidently different. If anyone has a clue on this (pointer to a psychoacoustic paper, or similar), je suis preneur :slight_smile:

As I understand it, the theory seems to be that we discriminate vowels based on the relationhship betwteen formant frequencies, with F2 being most important, then F1.

https://www.academia.edu/download/48078361/Acoustic_characteristics_of_American_Eng20160815-11238-1dvk2z2.pdf

I don’t have access to this one, but it seems like it tries to answer your question: https://asa.scitation.org/doi/abs/10.1121/1.421264

Thank you @weefuzzy, as always!

(The link does not work for me, though: “Oops! It looks like you’re in the wrong aisle”)

Weird. Does this one work?

(unpaywall FTW)

2 Likes

I am not 100% sure but I think the "the famous “aeiou” is from Gerhard Eckel and Francisco Iovino, 1994 IRCAM. I put them in a coll. I love the fact it comes from that era. I wanted to share it beyond the fact these could also be useful as initial theoretical machine learning material.

3 Likes

There is a good point here: at one moment of a production, one will have to decide which approach works best at minimized cpu costs.

It would be interesting to see if a KNN Classifier is cheaper than a MLP Classifier. I guess everything depends on how much is needed around each objects.
Depending on the needs, the learning process cost should only rarely count in that comparison.

Yes, it does :slight_smile:

thanks!