update

.gitignore

@@ -13,3 +13,4 @@ annot_distrib.pdf
 annot_distrib.tex
 humpback/annot
 humpback_CARIMAM/
+dolphin/zips

dolphin/dolphin.csv

+fn,label,dur,fs,pos
+FB79/85FB79_19.wav,FB79,1.49865625,96000,0.749328125
+FB79/85FB79_16.wav,FB79,1.50734375,96000,0.753671875
+FB79/85FB79_14.wav,FB79,1.50734375,96000,0.753671875
+FB79/85FB79_12.wav,FB79,1.50734375,96000,0.753671875
+FB79/85FB79_07.wav,FB79,1.50734375,96000,0.753671875
+FB79/85FB79_17.wav,FB79,1.515125,96000,0.7575625
+FB79/85FB79_03.wav,FB79,1.50734375,96000,0.753671875
+FB79/85FB79_08.wav,FB79,1.50734375,96000,0.753671875
+FB79/85FB79_04.wav,FB79,1.50734375,96000,0.753671875
+FB79/85FB79_05.wav,FB79,1.4799375,96000,0.73996875
+FB79/85FB79_20.wav,FB79,1.515125,96000,0.7575625
+FB79/85FB79_09.wav,FB79,1.50734375,96000,0.753671875
+FB79/85FB79_11.wav,FB79,1.50734375,96000,0.753671875
+FB79/85FB79_18.wav,FB79,1.49865625,96000,0.749328125
+FB79/85FB79_02.wav,FB79,1.50734375,96000,0.753671875
+FB79/85FB79_10.wav,FB79,1.50734375,96000,0.753671875
+FB79/85FB79_01.wav,FB79,1.50734375,96000,0.753671875
+FB79/85FB79_13.wav,FB79,1.50734375,96000,0.753671875
+FB79/85FB79_06.wav,FB79,1.53475,96000,0.767375
+FB79/85FB79_15.wav,FB79,1.50734375,96000,0.753671875
+FB25/89FB25_13.wav,FB25,2.0085,96000,1.00425
+FB25/89FB25_06.wav,FB25,2.0085,96000,1.00425
+FB25/89FB25_17.wav,FB25,2.0085,96000,1.00425
+FB25/89FB25_12.wav,FB25,2.0085,96000,1.00425
+FB25/89FB25_20.wav,FB25,2.0085,96000,1.00425
+FB25/89FB25_11.wav,FB25,2.0085,96000,1.00425
+FB25/89FB25_05.wav,FB25,2.0085,96000,1.00425
+FB25/89FB25_08.wav,FB25,2.0085,96000,1.00425
+FB25/89FB25_04.wav,FB25,2.0085,96000,1.00425
+FB25/89FB25_02.wav,FB25,2.0085,96000,1.00425
+FB25/89FB25_07.wav,FB25,2.0085,96000,1.00425
+FB25/89FB25_09.wav,FB25,2.0085,96000,1.00425
+FB25/89FB25_03.wav,FB25,2.0085,96000,1.00425
+FB25/89FB25_15.wav,FB25,2.0085,96000,1.00425
+FB25/89FB25_10.wav,FB25,2.0085,96000,1.00425
+FB25/89FB25_19.wav,FB25,2.0085,96000,1.00425
+FB25/89FB25_16.wav,FB25,2.0085,96000,1.00425
+FB25/89FB25_14.wav,FB25,2.0085,96000,1.00425
+FB25/89FB25_18.wav,FB25,2.0085,96000,1.00425
+FB25/89FB25_01.wav,FB25,2.0085,96000,1.00425
+FB02/92FB2_17.wav,FB02,1.51025,96000,0.755125
+FB02/92FB2_06.wav,FB02,1.50875,96000,0.754375
+FB02/92FB2_13.wav,FB02,1.51025,96000,0.755125
+FB02/92FB2_11.wav,FB02,1.50875,96000,0.754375
+FB02/92FB2_15.wav,FB02,1.51025,96000,0.755125
+FB02/92FB2_12.wav,FB02,1.50875,96000,0.754375
+FB02/92FB2_10.wav,FB02,1.50875,96000,0.754375
+FB02/92FB2_16.wav,FB02,1.5282291666666667,96000,0.7641145833333334
+FB02/92FB2_14.wav,FB02,1.4922708333333334,96000,0.7461354166666667
+FB02/92FB2_08.wav,FB02,1.50875,96000,0.754375
+FB02/92FB2_07.wav,FB02,1.50875,96000,0.754375
+FB02/92FB2_18.wav,FB02,1.4922708333333334,96000,0.7461354166666667
+FB02/92FB2_20.wav,FB02,1.51025,96000,0.755125
+FB02/92FB2_01.wav,FB02,1.50875,96000,0.754375
+FB02/92FB2_19.wav,FB02,1.4922708333333334,96000,0.7461354166666667
+FB02/92FB2_04.wav,FB02,1.4865625,96000,0.74328125
+FB02/92FB2_02.wav,FB02,1.50875,96000,0.754375
+FB02/92FB2_09.wav,FB02,1.50875,96000,0.754375
+FB02/92FB2_05.wav,FB02,1.50875,96000,0.754375
+FB02/92FB2_03.wav,FB02,1.50875,96000,0.754375
+FB92/89FB92_03.wav,FB92,2.003739583333333,96000,1.0018697916666666
+FB92/89FB92_06.wav,FB92,2.003739583333333,96000,1.0018697916666666
+FB92/89FB92_19.wav,FB92,2.003739583333333,96000,1.0018697916666666
+FB92/89FB92_13.wav,FB92,2.003739583333333,96000,1.0018697916666666
+FB92/89FB92_15.wav,FB92,2.003739583333333,96000,1.0018697916666666
+FB92/89FB92_02.wav,FB92,2.003739583333333,96000,1.0018697916666666
+FB92/89FB92_09.wav,FB92,2.003739583333333,96000,1.0018697916666666
+FB92/89FB92_10.wav,FB92,2.003739583333333,96000,1.0018697916666666
+FB92/89FB92_04.wav,FB92,2.003739583333333,96000,1.0018697916666666
+FB92/89FB92_20.wav,FB92,2.0171875,96000,1.00859375
+FB92/89FB92_01.wav,FB92,2.003739583333333,96000,1.0018697916666666
+FB92/89FB92_11.wav,FB92,2.003739583333333,96000,1.0018697916666666
+FB92/89FB92_18.wav,FB92,2.003739583333333,96000,1.0018697916666666
+FB92/89FB92_08.wav,FB92,2.003739583333333,96000,1.0018697916666666
+FB92/89FB92_05.wav,FB92,2.003739583333333,96000,1.0018697916666666
+FB92/89FB92_14.wav,FB92,1.9902916666666666,96000,0.9951458333333333
+FB92/89FB92_17.wav,FB92,2.003739583333333,96000,1.0018697916666666
+FB92/89FB92_12.wav,FB92,2.003739583333333,96000,1.0018697916666666
+FB92/89FB92_16.wav,FB92,2.003739583333333,96000,1.0018697916666666
+FB92/89FB92_07.wav,FB92,2.003739583333333,96000,1.0018697916666666
+FB09/85FB9_02.wav,FB09,1.97834375,96000,0.989171875
+FB09/85FB9_10.wav,FB09,2.0010833333333333,96000,1.0005416666666667
+FB09/85FB9_13.wav,FB09,2.0010833333333333,96000,1.0005416666666667
+FB09/85FB9_08.wav,FB09,2.0010833333333333,96000,1.0005416666666667
+FB09/85FB9_07.wav,FB09,2.0010833333333333,96000,1.0005416666666667
+FB09/85FB9_05.wav,FB09,2.0010833333333333,96000,1.0005416666666667
+FB09/85FB9_14.wav,FB09,2.0010833333333333,96000,1.0005416666666667
+FB09/85FB9_09.wav,FB09,2.0010833333333333,96000,1.0005416666666667
+FB09/85FB9_20.wav,FB09,2.0010833333333333,96000,1.0005416666666667
+FB09/85FB9_11.wav,FB09,2.0010833333333333,96000,1.0005416666666667
+FB09/85FB9_15.wav,FB09,2.0010833333333333,96000,1.0005416666666667
+FB09/85FB9_17.wav,FB09,2.0010833333333333,96000,1.0005416666666667
+FB09/85FB9_01.wav,FB09,2.0010833333333333,96000,1.0005416666666667
+FB09/85FB9_06.wav,FB09,2.0010833333333333,96000,1.0005416666666667
+FB09/85FB9_03.wav,FB09,2.0010833333333333,96000,1.0005416666666667
+FB09/85FB9_12.wav,FB09,2.0010833333333333,96000,1.0005416666666667
+FB09/85FB9_19.wav,FB09,2.0010833333333333,96000,1.0005416666666667
+FB09/85FB9_18.wav,FB09,2.0010833333333333,96000,1.0005416666666667
+FB09/85FB9_16.wav,FB09,2.0010833333333333,96000,1.0005416666666667
+FB09/85FB9_04.wav,FB09,2.0010833333333333,96000,1.0005416666666667
+FB07/87FB7_14.wav,FB07,2.0063541666666667,96000,1.0031770833333333
+FB07/87FB7_13.wav,FB07,2.0063541666666667,96000,1.0031770833333333
+FB07/87FB7_20.wav,FB07,2.0063541666666667,96000,1.0031770833333333
+FB07/87FB7_18.wav,FB07,1.9881145833333334,96000,0.9940572916666667
+FB07/87FB7_07.wav,FB07,2.0063541666666667,96000,1.0031770833333333
+FB07/87FB7_05.wav,FB07,2.0063541666666667,96000,1.0031770833333333
+FB07/87FB7_11.wav,FB07,2.0063541666666667,96000,1.0031770833333333
+FB07/87FB7_02.wav,FB07,2.0063541666666667,96000,1.0031770833333333
+FB07/87FB7_08.wav,FB07,1.9881145833333334,96000,0.9940572916666667
+FB07/87FB7_09.wav,FB07,2.0063541666666667,96000,1.0031770833333333
+FB07/87FB7_15.wav,FB07,2.0063541666666667,96000,1.0031770833333333
+FB07/87FB7_12.wav,FB07,2.0063541666666667,96000,1.0031770833333333
+FB07/87FB7_01.wav,FB07,2.0063541666666667,96000,1.0031770833333333
+FB07/87FB7_06.wav,FB07,2.0063541666666667,96000,1.0031770833333333
+FB07/87FB7_10.wav,FB07,1.9881145833333334,96000,0.9940572916666667
+FB07/87FB7_16.wav,FB07,2.0063541666666667,96000,1.0031770833333333
+FB07/87FB7_19.wav,FB07,2.0063541666666667,96000,1.0031770833333333
+FB07/87FB7_04.wav,FB07,2.0063541666666667,96000,1.0031770833333333
+FB07/87FB7_17.wav,FB07,2.0063541666666667,96000,1.0031770833333333
+FB07/87FB7_03.wav,FB07,2.0063541666666667,96000,1.0031770833333333
+FB35/89FB35_11.wav,FB35,2.002,96000,1.001
+FB35/89FB35_02.wav,FB35,1.9970833333333333,96000,0.9985416666666667
+FB35/89FB35_03.wav,FB35,1.9970833333333333,96000,0.9985416666666667
+FB35/89FB35_06.wav,FB35,1.9970833333333333,96000,0.9985416666666667
+FB35/89FB35_07.wav,FB35,1.9970833333333333,96000,0.9985416666666667
+FB35/89FB35_16.wav,FB35,2.002,96000,1.001
+FB35/89FB35_04.wav,FB35,1.9970833333333333,96000,0.9985416666666667
+FB35/89FB35_13.wav,FB35,2.002,96000,1.001
+FB35/89FB35_10.wav,FB35,1.9970833333333333,96000,0.9985416666666667
+FB35/89FB35_17.wav,FB35,2.002,96000,1.001
+FB35/89FB35_18.wav,FB35,2.002,96000,1.001
+FB35/89FB35_01.wav,FB35,1.9970833333333333,96000,0.9985416666666667
+FB35/89FB35_15.wav,FB35,2.002,96000,1.001
+FB35/89FB35_05.wav,FB35,1.9970833333333333,96000,0.9985416666666667
+FB35/89FB35_20.wav,FB35,2.002,96000,1.001
+FB35/89FB35_19.wav,FB35,2.002,96000,1.001
+FB35/89FB35_09.wav,FB35,1.9970833333333333,96000,0.9985416666666667
+FB35/89FB35_08.wav,FB35,1.9970833333333333,96000,0.9985416666666667
+FB35/89FB35_12.wav,FB35,2.002,96000,1.001
+FB35/89FB35_14.wav,FB35,2.002,96000,1.001
+FB55/02FB55_14.wav,FB55,2.010544217687075,88200,1.0052721088435375
+FB55/02FB55_20.wav,FB55,2.010544217687075,88200,1.0052721088435375
+FB55/02FB55_15.wav,FB55,2.010544217687075,88200,1.0052721088435375
+FB55/02FB55_16.wav,FB55,2.010544217687075,88200,1.0052721088435375
+FB55/02FB55_04.wav,FB55,1.9930612244897958,88200,0.9965306122448979
+FB55/02FB55_11.wav,FB55,2.010544217687075,88200,1.0052721088435375
+FB55/02FB55_18.wav,FB55,1.9930612244897958,88200,0.9965306122448979
+FB55/02FB55_13.wav,FB55,2.010544217687075,88200,1.0052721088435375
+FB55/02FB55_10.wav,FB55,2.028027210884354,88200,1.014013605442177
+FB55/02FB55_03.wav,FB55,2.010544217687075,88200,1.0052721088435375
+FB55/02FB55_05.wav,FB55,2.010544217687075,88200,1.0052721088435375
+FB55/02FB55_08.wav,FB55,2.010544217687075,88200,1.0052721088435375
+FB55/02FB55_06.wav,FB55,2.010544217687075,88200,1.0052721088435375
+FB55/02FB55_19.wav,FB55,2.010544217687075,88200,1.0052721088435375
+FB55/02FB55_07.wav,FB55,2.010544217687075,88200,1.0052721088435375
+FB55/02FB55_02.wav,FB55,1.9930612244897958,88200,0.9965306122448979
+FB55/02FB55_01.wav,FB55,1.9930612244897958,88200,0.9965306122448979
+FB55/02FB55_09.wav,FB55,1.975578231292517,88200,0.9877891156462585
+FB55/02FB55_12.wav,FB55,2.010544217687075,88200,1.0052721088435375
+FB55/02FB55_17.wav,FB55,2.010544217687075,88200,1.0052721088435375
+FB33/84FB33_01.wav,FB33,2.491125,96000,1.2455625
+FB33/84FB33_05.wav,FB33,2.5108958333333335,96000,1.2554479166666668
+FB33/84FB33_18.wav,FB33,2.4934375,96000,1.24671875
+FB33/84FB33_20.wav,FB33,2.518125,96000,1.2590625
+FB33/84FB33_17.wav,FB33,2.4934375,96000,1.24671875
+FB33/84FB33_16.wav,FB33,2.4934375,96000,1.24671875
+FB33/84FB33_19.wav,FB33,2.4934375,96000,1.24671875
+FB33/84FB33_02.wav,FB33,2.5108958333333335,96000,1.2554479166666668
+FB33/84FB33_03.wav,FB33,2.491125,96000,1.2455625
+FB33/84FB33_04.wav,FB33,2.5108958333333335,96000,1.2554479166666668
+FB33/84FB33_08.wav,FB33,2.491125,96000,1.2455625
+FB33/84FB33_14.wav,FB33,2.518125,96000,1.2590625
+FB33/84FB33_13.wav,FB33,2.518125,96000,1.2590625
+FB33/84FB33_09.wav,FB33,2.491125,96000,1.2455625
+FB33/84FB33_15.wav,FB33,2.50578125,96000,1.252890625
+FB33/84FB33_06.wav,FB33,2.5108958333333335,96000,1.2554479166666668
+FB33/84FB33_07.wav,FB33,2.491125,96000,1.2455625
+FB33/84FB33_11.wav,FB33,2.491125,96000,1.2455625
+FB33/84FB33_12.wav,FB33,2.491125,96000,1.2455625
+FB33/84FB33_10.wav,FB33,2.5108958333333335,96000,1.2554479166666668
+FB101/00FB101_01.wav,FB101,2.0003125,96000,1.00015625
+FB101/00FB101_08.wav,FB101,2.0003125,96000,1.00015625
+FB101/00FB101_07.wav,FB101,2.0003125,96000,1.00015625
+FB101/00FB101_10.wav,FB101,2.0003125,96000,1.00015625
+FB101/00FB101_02.wav,FB101,2.0003125,96000,1.00015625
+FB101/00FB101_14.wav,FB101,2.0183333333333335,96000,1.0091666666666668
+FB101/00FB101_11.wav,FB101,1.9822916666666666,96000,0.9911458333333333
+FB101/00FB101_04.wav,FB101,2.0003125,96000,1.00015625
+FB101/00FB101_18.wav,FB101,2.0003125,96000,1.00015625
+FB101/00FB101_06.wav,FB101,2.0003125,96000,1.00015625
+FB101/00FB101_16.wav,FB101,2.0003125,96000,1.00015625
+FB101/00FB101_13.wav,FB101,2.0003125,96000,1.00015625
+FB101/00FB101_09.wav,FB101,2.0003125,96000,1.00015625
+FB101/00FB101_05.wav,FB101,2.0003125,96000,1.00015625
+FB101/00FB101_12.wav,FB101,2.0003125,96000,1.00015625
+FB101/00FB101_03.wav,FB101,2.0003125,96000,1.00015625
+FB101/00FB101_20.wav,FB101,2.0003125,96000,1.00015625
+FB101/00FB101_17.wav,FB101,2.0003125,96000,1.00015625
+FB101/00FB101_15.wav,FB101,2.0003125,96000,1.00015625
+FB101/00FB101_19.wav,FB101,2.0003125,96000,1.00015625
+FB05/89FB5_14.wav,FB05,1.9953333333333334,96000,0.9976666666666667
+FB05/89FB5_02.wav,FB05,1.9970416666666666,96000,0.9985208333333333
+FB05/89FB5_01.wav,FB05,1.9982083333333334,96000,0.9991041666666667
+FB05/89FB5_17.wav,FB05,1.9953333333333334,96000,0.9976666666666667
+FB05/89FB5_19.wav,FB05,1.9953333333333334,96000,0.9976666666666667
+FB05/89FB5_04.wav,FB05,1.9970416666666666,96000,0.9985208333333333
+FB05/89FB5_10.wav,FB05,1.9953333333333334,96000,0.9976666666666667
+FB05/89FB5_16.wav,FB05,1.9953333333333334,96000,0.9976666666666667
+FB05/89FB5_06.wav,FB05,1.9970416666666666,96000,0.9985208333333333
+FB05/89FB5_03.wav,FB05,1.9970416666666666,96000,0.9985208333333333
+FB05/89FB5_15.wav,FB05,1.9953333333333334,96000,0.9976666666666667
+FB05/89FB5_11.wav,FB05,1.9953333333333334,96000,0.9976666666666667
+FB05/89FB5_12.wav,FB05,1.9953333333333334,96000,0.9976666666666667
+FB05/89FB5_08.wav,FB05,1.9953333333333334,96000,0.9976666666666667
+FB05/89FB5_07.wav,FB05,2.0213958333333335,96000,1.0106979166666668
+FB05/89FB5_20.wav,FB05,1.9953333333333334,96000,0.9976666666666667
+FB05/89FB5_18.wav,FB05,1.9953333333333334,96000,0.9976666666666667
+FB05/89FB5_13.wav,FB05,1.9466666666666668,96000,0.9733333333333334
+FB05/89FB5_09.wav,FB05,2.0196666666666667,96000,1.0098333333333334
+FB05/89FB5_05.wav,FB05,1.9970416666666666,96000,0.9985208333333333
+FB182/90FB182_02.wav,FB182,2.5000104166666666,96000,1.2500052083333333
+FB182/90FB182_04.wav,FB182,2.5000104166666666,96000,1.2500052083333333
+FB182/90FB182_03.wav,FB182,2.5000104166666666,96000,1.2500052083333333
+FB182/90FB182_11.wav,FB182,2.5000104166666666,96000,1.2500052083333333
+FB182/90FB182_17.wav,FB182,2.5000104166666666,96000,1.2500052083333333
+FB182/90FB182_08.wav,FB182,2.5000104166666666,96000,1.2500052083333333
+FB182/90FB182_16.wav,FB182,2.5000104166666666,96000,1.2500052083333333
+FB182/90FB182_19.wav,FB182,2.5000104166666666,96000,1.2500052083333333
+FB182/90FB182_13.wav,FB182,2.5000104166666666,96000,1.2500052083333333
+FB182/90FB182_06.wav,FB182,2.5000104166666666,96000,1.2500052083333333
+FB182/90FB182_07.wav,FB182,2.5000104166666666,96000,1.2500052083333333
+FB182/90FB182_10.wav,FB182,2.5000104166666666,96000,1.2500052083333333
+FB182/90FB182_15.wav,FB182,2.5000104166666666,96000,1.2500052083333333
+FB182/90FB182_01.wav,FB182,2.5000104166666666,96000,1.2500052083333333
+FB182/90FB182_18.wav,FB182,2.5000104166666666,96000,1.2500052083333333
+FB182/90FB182_20.wav,FB182,2.5000104166666666,96000,1.2500052083333333
+FB182/90FB182_12.wav,FB182,2.5000104166666666,96000,1.2500052083333333
+FB182/F182-2005-SW-IND10622.wav,FB182,1.4600104166666668,96000,0.7300052083333334
+FB182/90FB182_05.wav,FB182,2.5000104166666666,96000,1.2500052083333333
+FB182/90FB182_09.wav,FB182,2.5000104166666666,96000,1.2500052083333333
+FB122/94FB122_04.wav,FB122,2.0151041666666667,96000,1.0075520833333333
+FB122/94FB122_13.wav,FB122,1.99609375,96000,0.998046875
+FB122/94FB122_18.wav,FB122,2.001333333333333,96000,1.0006666666666666
+FB122/94FB122_19.wav,FB122,2.001333333333333,96000,1.0006666666666666
+FB122/94FB122_15.wav,FB122,2.0151041666666667,96000,1.0075520833333333
+FB122/94FB122_08.wav,FB122,2.0151041666666667,96000,1.0075520833333333
+FB122/94FB122_03.wav,FB122,2.0151041666666667,96000,1.0075520833333333
+FB122/94FB122_02.wav,FB122,1.99609375,96000,0.998046875
+FB122/94FB122_10.wav,FB122,2.0151041666666667,96000,1.0075520833333333
+FB122/94FB122_11.wav,FB122,1.99609375,96000,0.998046875
+FB122/94FB122_17.wav,FB122,2.0145,96000,1.00725
+FB122/94FB122_06.wav,FB122,1.99609375,96000,0.998046875
+FB122/94FB122_12.wav,FB122,2.0151041666666667,96000,1.0075520833333333
+FB122/94FB122_05.wav,FB122,1.99609375,96000,0.998046875
+FB122/94FB122_16.wav,FB122,2.0145,96000,1.00725
+FB122/94FB122_07.wav,FB122,2.0151041666666667,96000,1.0075520833333333
+FB122/94FB122_20.wav,FB122,2.001333333333333,96000,1.0006666666666666
+FB122/94FB122_14.wav,FB122,1.99609375,96000,0.998046875
+FB122/94FB122_01.wav,FB122,1.99609375,96000,0.998046875
+FB122/94FB122_09.wav,FB122,2.0151041666666667,96000,1.0075520833333333
+FB10/85FB10_05.wav,FB10,2.5017708333333335,96000,1.2508854166666667
+FB10/85FB10_06.wav,FB10,2.46603125,96000,1.233015625
+FB10/85FB10_08.wav,FB10,2.5017708333333335,96000,1.2508854166666667
+FB10/85FB10_20.wav,FB10,2.5017708333333335,96000,1.2508854166666667
+FB10/85FB10_18.wav,FB10,2.5017708333333335,96000,1.2508854166666667
+FB10/85FB10_10.wav,FB10,2.5017708333333335,96000,1.2508854166666667
+FB10/85FB10_14.wav,FB10,2.5017708333333335,96000,1.2508854166666667
+FB10/85FB10_12.wav,FB10,2.5017708333333335,96000,1.2508854166666667
+FB10/85FB10_17.wav,FB10,2.5017708333333335,96000,1.2508854166666667
+FB10/85FB10_13.wav,FB10,2.5017708333333335,96000,1.2508854166666667
+FB10/85FB10_09.wav,FB10,2.5017708333333335,96000,1.2508854166666667
+FB10/85FB10_01.wav,FB10,2.5017708333333335,96000,1.2508854166666667
+FB10/85FB10_07.wav,FB10,2.5017708333333335,96000,1.2508854166666667
+FB10/85FB10_11.wav,FB10,2.5017708333333335,96000,1.2508854166666667
+FB10/85FB10_02.wav,FB10,2.5017708333333335,96000,1.2508854166666667
+FB10/85FB10_04.wav,FB10,2.5017708333333335,96000,1.2508854166666667
+FB10/85FB10_19.wav,FB10,2.5017708333333335,96000,1.2508854166666667
+FB10/85FB10_03.wav,FB10,2.5017708333333335,96000,1.2508854166666667
+FB10/85FB10_16.wav,FB10,2.5017708333333335,96000,1.2508854166666667
+FB10/85FB10_15.wav,FB10,2.5017708333333335,96000,1.2508854166666667
+FB11/87FB11_08.wav,FB11,2.0,96000,1.0
+FB11/87FB11_17.wav,FB11,2.0,96000,1.0
+FB11/87FB11_16.wav,FB11,2.013333333333333,96000,1.0066666666666666
+FB11/87FB11_18.wav,FB11,2.0,96000,1.0
+FB11/87FB11_05.wav,FB11,1.9976041666666666,96000,0.9988020833333333
+FB11/87FB11_01.wav,FB11,2.0094583333333333,96000,1.0047291666666667
+FB11/87FB11_11.wav,FB11,2.0,96000,1.0
+FB11/87FB11_03.wav,FB11,2.013333333333333,96000,1.0066666666666666
+FB11/87FB11_10.wav,FB11,2.0,96000,1.0
+FB11/87FB11_19.wav,FB11,2.0,96000,1.0
+FB11/87FB11_04.wav,FB11,1.9976041666666666,96000,0.9988020833333333
+FB11/87FB11_07.wav,FB11,2.013333333333333,96000,1.0066666666666666
+FB11/87FB11_09.wav,FB11,2.0,96000,1.0
+FB11/87FB11_02.wav,FB11,1.9976041666666666,96000,0.9988020833333333
+FB11/87FB11_13.wav,FB11,1.9866666666666666,96000,0.9933333333333333
+FB11/87FB11_14.wav,FB11,2.013333333333333,96000,1.0066666666666666
+FB11/87FB11_12.wav,FB11,2.0,96000,1.0
+FB11/87FB11_20.wav,FB11,2.0,96000,1.0
+FB11/87FB11_15.wav,FB11,2.0,96000,1.0
+FB11/87FB11_06.wav,FB11,1.9976041666666666,96000,0.9988020833333333
+FB131/92FB131_12.wav,FB131,2.4680045351473923,88200,1.2340022675736961
+FB131/92FB131_19.wav,FB131,2.5068707482993196,88200,1.2534353741496598
+FB131/92FB131_03.wav,FB131,2.5068707482993196,88200,1.2534353741496598
+FB131/92FB131_09.wav,FB131,2.5068707482993196,88200,1.2534353741496598
+FB131/92FB131_07.wav,FB131,2.5068707482993196,88200,1.2534353741496598
+FB131/92FB131_11.wav,FB131,2.5068707482993196,88200,1.2534353741496598
+FB131/92FB131_20.wav,FB131,2.5068707482993196,88200,1.2534353741496598
+FB131/92FB131_16.wav,FB131,2.5068707482993196,88200,1.2534353741496598
+FB131/92FB131_05.wav,FB131,2.5068707482993196,88200,1.2534353741496598
+FB131/92FB131_13.wav,FB131,2.5068707482993196,88200,1.2534353741496598
+FB131/92FB131_02.wav,FB131,2.5068707482993196,88200,1.2534353741496598
+FB131/92FB131_10.wav,FB131,2.5068707482993196,88200,1.2534353741496598
+FB131/92FB131_04.wav,FB131,2.5068707482993196,88200,1.2534353741496598
+FB131/92FB131_15.wav,FB131,2.5068707482993196,88200,1.2534353741496598
+FB131/92FB131_18.wav,FB131,2.5068707482993196,88200,1.2534353741496598
+FB131/92FB131_01.wav,FB131,2.5068707482993196,88200,1.2534353741496598
+FB131/92FB131_08.wav,FB131,2.5068707482993196,88200,1.2534353741496598
+FB131/92FB131_14.wav,FB131,2.5068707482993196,88200,1.2534353741496598
+FB131/92FB131_17.wav,FB131,2.5068707482993196,88200,1.2534353741496598
+FB131/95FB131E.WAV,FB131,3.7029,80000,1.85145
+FB15/91FB15_20.wav,FB15,2.0120416666666667,96000,1.0060208333333334
+FB15/91FB15_02.wav,FB15,2.0120416666666667,96000,1.0060208333333334
+FB15/91FB15_09.wav,FB15,2.0120416666666667,96000,1.0060208333333334
+FB15/91FB15_11.wav,FB15,2.0120416666666667,96000,1.0060208333333334
+FB15/91FB15_03.wav,FB15,2.0120416666666667,96000,1.0060208333333334
+FB15/91FB15_06.wav,FB15,1.9886458333333332,96000,0.9943229166666666
+FB15/91FB15_07.wav,FB15,2.0120416666666667,96000,1.0060208333333334
+FB15/91FB15_05.wav,FB15,2.0120416666666667,96000,1.0060208333333334
+FB15/91FB15_12.wav,FB15,2.0120416666666667,96000,1.0060208333333334
+FB15/91FB15_19.wav,FB15,2.0120416666666667,96000,1.0060208333333334
+FB15/91FB15_10.wav,FB15,2.0354375,96000,1.01771875
+FB15/91FB15_16.wav,FB15,2.0120416666666667,96000,1.0060208333333334
+FB15/91FB15_13.wav,FB15,2.0120416666666667,96000,1.0060208333333334
+FB15/91FB15_14.wav,FB15,2.0120416666666667,96000,1.0060208333333334
+FB15/91FB15_18.wav,FB15,2.0120416666666667,96000,1.0060208333333334
+FB15/91FB15_15.wav,FB15,2.0120416666666667,96000,1.0060208333333334
+FB15/91FB15_04.wav,FB15,2.0120416666666667,96000,1.0060208333333334
+FB15/91FB15_17.wav,FB15,2.0120416666666667,96000,1.0060208333333334
+FB15/91FB15_08.wav,FB15,2.0120416666666667,96000,1.0060208333333334
+FB15/91FB15_01.wav,FB15,2.0120416666666667,96000,1.0060208333333334
+FB71/90FB71_12.wav,FB71,2.003625,96000,1.0018125
+FB71/90FB71_04.wav,FB71,2.00625,96000,1.003125
+FB71/90FB71_20.wav,FB71,2.01825,96000,1.009125
+FB71/90FB71_18.wav,FB71,2.003625,96000,1.0018125
+FB71/90FB71_08.wav,FB71,2.0048958333333333,96000,1.0024479166666667
+FB71/90FB71_13.wav,FB71,2.003625,96000,1.0018125
+FB71/90FB71_02.wav,FB71,2.00625,96000,1.003125
+FB71/90FB71_11.wav,FB71,2.0048958333333333,96000,1.0024479166666667
+FB71/90FB71_06.wav,FB71,2.0048958333333333,96000,1.0024479166666667
+FB71/90FB71_17.wav,FB71,2.003625,96000,1.0018125
+FB71/90FB71_16.wav,FB71,2.003625,96000,1.0018125
+FB71/FB71A.WAV,FB71,1.2203375,80000,0.61016875
+FB71/90FB71_15.wav,FB71,2.003625,96000,1.0018125
+FB71/90FB71_14.wav,FB71,2.003625,96000,1.0018125
+FB71/90FB71_07.wav,FB71,2.0048958333333333,96000,1.0024479166666667
+FB71/90FB71_19.wav,FB71,2.003625,96000,1.0018125
+FB71/90FB71_05.wav,FB71,2.00625,96000,1.003125
+FB71/90FB71_10.wav,FB71,2.0048958333333333,96000,1.0024479166666667
+FB71/90FB71_09.wav,FB71,2.0048958333333333,96000,1.0024479166666667
+FB71/90FB71_03.wav,FB71,2.00625,96000,1.003125
+FB67/89FB67_12.wav,FB67,2.99975,96000,1.499875
+FB67/89FB67_10.wav,FB67,3.0138333333333334,96000,1.5069166666666667
+FB67/89FB67_20.wav,FB67,2.99975,96000,1.499875
+FB67/89FB67_05.wav,FB67,2.99975,96000,1.499875
+FB67/89FB67_17.wav,FB67,3.0138333333333334,96000,1.5069166666666667
+FB67/89FB67_19.wav,FB67,2.99975,96000,1.499875
+FB67/89FB67_08.wav,FB67,3.0138333333333334,96000,1.5069166666666667
+FB67/89FB67_04.wav,FB67,3.0138333333333334,96000,1.5069166666666667
+FB67/89FB67_14.wav,FB67,3.0138333333333334,96000,1.5069166666666667
+FB67/89FB67_02.wav,FB67,2.9856666666666665,96000,1.4928333333333332
+FB67/89FB67_01.wav,FB67,3.0138333333333334,96000,1.5069166666666667
+FB67/89FB67_16.wav,FB67,2.99975,96000,1.499875
+FB67/89FB67_18.wav,FB67,3.0138333333333334,96000,1.5069166666666667
+FB67/89FB67_03.wav,FB67,2.99975,96000,1.499875
+FB67/89FB67_07.wav,FB67,3.0138333333333334,96000,1.5069166666666667
+FB67/89FB67_09.wav,FB67,3.0138333333333334,96000,1.5069166666666667
+FB67/89FB67_11.wav,FB67,3.0138333333333334,96000,1.5069166666666667
+FB67/89FB67_13.wav,FB67,3.0138333333333334,96000,1.5069166666666667
+FB67/89FB67_06.wav,FB67,2.99975,96000,1.499875
+FB67/89FB67_15.wav,FB67,3.0279166666666666,96000,1.5139583333333333
+FB163/91FB163_20.wav,FB163,2.49903125,96000,1.249515625
+FB163/91FB163_06.wav,FB163,2.510875,96000,1.2554375
+FB163/FB163F_adjSR.wav,FB163,0.9566,80000,0.4783
+FB163/FB163G_adjSR.wav,FB163,1.5,80000,0.75
+FB163/91FB163_10.wav,FB163,2.49903125,96000,1.249515625
+FB163/91FB163_07.wav,FB163,2.510875,96000,1.2554375
+FB163/91FB163_01.wav,FB163,2.5108333333333333,96000,1.2554166666666666
+FB163/91FB163_02.wav,FB163,2.489,96000,1.2445
+FB163/91FB163_11.wav,FB163,2.510875,96000,1.2554375
+FB163/91FB163_12.wav,FB163,2.49903125,96000,1.249515625
+FB163/91FB163_13.wav,FB163,2.510875,96000,1.2554375
+FB163/91FB163_15.wav,FB163,2.49903125,96000,1.249515625
+FB163/91FB163_19.wav,FB163,2.49903125,96000,1.249515625
+FB163/91FB163_09.wav,FB163,2.510875,96000,1.2554375
+FB163/FB163K_adjSR.wav,FB163,1.3607875,80000,0.68039375
+FB163/FB163D_adjSR.wav,FB163,1.8088125,80000,0.90440625
+FB163/91FB163_04.wav,FB163,2.510875,96000,1.2554375
+FB163/91FB163_03.wav,FB163,2.5108333333333333,96000,1.2554166666666666
+FB163/91FB163_18.wav,FB163,2.510875,96000,1.2554375
+FB163/91FB163_16.wav,FB163,2.49903125,96000,1.249515625

good_species.txt

@@ -4,4 +4,5 @@ california_thrashers
 cassin_vireo
 black-headed_grosbeaks
 humpback
-zebra_finch
+dolphin
+otter

models.py

@@ -41,12 +41,12 @@ meta = {
     'sampleDur': 2
   },
   'humpback':{
-    'nfft': 512,
+    'nfft': 1024,
     'sr': 11025,
     'sampleDur': 2
   },
   'humpback2':{
-    'nfft': 512,
+    'nfft': 2048,
     'sr': 11025,
     'sampleDur': 2
   },
@@ -57,12 +57,17 @@ meta = {
   },
   'otter':{
     'nfft':2048,
-    'sr':48000,
-    'sampleDur':0.5
+    'sr':96000,
+    'sampleDur':1
+  },
+  'dolphin':{
+    'nfft':512,
+    'sr':96000,
+    'sampleDur':2
   }
 }
 
-vgg16 = torchmodels.vgg16(pretrained=True) # weights=torchmodels.VGG16_Weights.DEFAULT)
+vgg16 = torchmodels.vgg16(weights=torchmodels.VGG16_Weights.DEFAULT)
 vgg16 = vgg16.features[:13]
 for nm, mod in vgg16.named_modules():
     if isinstance(mod, nn.MaxPool2d):
@@ -82,6 +87,13 @@ frontend = {
     nn.InstanceNorm2d(1),
     u.Croper2D(n_mel, 128)
   ),
+  'logMel_vggish': lambda sr, nfft, sampleDur, n_mel : nn.Sequential(
+    STFT(nfft, int((sampleDur*sr - nfft)/96)),
+    MelFilter(sr, nfft, n_mel, 0, sr//2),
+    Log1p(7, trainable=False),
+    nn.InstanceNorm2d(1),
+    u.Croper2D(n_mel, 128)
+  ),
   'logSTFT': lambda sr, nfft, sampleDur, n_mel : nn.Sequential(
     STFT(nfft, int((sampleDur*sr - nfft)/128)),
     Log1p(7, trainable=False),
@@ -117,8 +129,28 @@ sparrow_VQ_encoder = lambda nfeat, shape : nn.Sequential(
   u.Reshape(nfeat * shape[0] * shape[1])
 )
 
+sparrow_encoder_maxPool = lambda nfeat, shape : nn.Sequential(
+  nn.Conv2d(1, 32, 3, bias=False, padding=1),
+  nn.BatchNorm2d(32),
+  nn.ReLU(True),
+  nn.Conv2d(32, 32, 3, bias=False, padding=1),
+  nn.BatchNorm2d(32),
+  nn.MaxPool2d((4, 4)),
+  nn.ReLU(True),
+  nn.Conv2d(32, 32, 3, bias=False, padding=1),
+  nn.BatchNorm2d(32),
+  nn.ReLU(True),
+  nn.Conv2d(32, 32, 3, bias=False, padding=1),
+  nn.BatchNorm2d(32),
+  nn.ReLU(True),
+  nn.Conv2d(32, nfeat, 3, bias=False, padding=1),
+  nn.MaxPool2d((4, 4)),
+  u.Reshape(nfeat * shape[0] * shape[1])
+)
+
+
 sparrow_encoder = lambda nfeat, shape : nn.Sequential(
-  nn.Conv2d(1, 32, 3, stride=2, bias=False, padding=(1)),
+  nn.Conv2d(1, 32, 3, stride=2, bias=False, padding=1),
   nn.BatchNorm2d(32),
   nn.ReLU(True),
   nn.Conv2d(32, 64, 3, stride=2, bias=False, padding=1),
@@ -177,7 +209,7 @@ sparrow_decoder = lambda nfeat, shape : nn.Sequential(
   nn.Conv2d(32, 1, (3, 3), bias=False, padding=1),
   nn.BatchNorm2d(1),
   nn.ReLU(True),
-  nn.Conv2d(1, 1, (3, 3), bias=False, padding=1),
+  nn.Conv2d(1, 1, (3, 3), bias=False, padding=1)
 )
 
 sparrow_decoder_old = lambda nfeat, shape : nn.Sequential(

plot_annot_distrib.py

@@ -2,6 +2,15 @@ import matplotlib.pyplot as plt
 import pandas as pd, numpy as np
 
 species = np.loadtxt('good_species.txt', dtype=str)
+info = {
+    'bengalese_finch1': ['bengalese finch', 'nicholson2017bengalese', 'bird'],
+    'bengalese_finch2': ['bengalese finch', 'koumura2016birdsongrecognition', 'bird'],
+    'california_thrashers': ['california trashers', 'arriaga2015bird', 'bird'],
+    'cassin_vireo': ['cassin vireo', 'arriaga2015bird', 'bird'],
+    'black-headed_grosbeaks': ['black-headed grosbeaks', 'arriaga2015bird', 'bird'],
+    'zebra_finch': ['zebra finch', 'elie2018zebra', 'bird'],
+    'humpback': ['humpback whale', 'malige2021use', 'cetacean'],
+}
 
 fig, ax = plt.subplots(nrows=4, ncols=3, figsize=(10, 10))
 for i, specie in enumerate(species):
@@ -12,10 +21,10 @@ plt.tight_layout()
 plt.savefig('annot_distrib.pdf')
 
 
-a = "Specie & \# Classes & \# Annotated samples & \# Samples & Proportion of annotations\\\\ \hline \n"
+a = "Specie & \# Classes & \# Samples & Annotations \% \\\\ \hline \n"
 for specie in species:
     df = pd.read_csv(f'{specie}/{specie}.csv')
-    a += f"{specie.replace('_',' ')} & {df.label.nunique()} & {(~df.label.isna()).sum()} & {len(df)} & {int(100*(~df.label.isna()).sum()/len(df))} \\\\ \hline \n"
+    a += f"{info[specie][0]} \cite{{{info[specie][1]}}} & {df.label.nunique()} & {len(df)} & {int(100*(~df.label.isna()).sum()/len(df))} \\\\ \hline \n"
 f = open('annot_distrib.tex', 'w')
 f.write(a)
 f.close()

plot_results_hdbcsan.py

@@ -5,7 +5,7 @@ import numpy as np
 from sklearn import metrics
 
 species = np.loadtxt('good_species.txt', dtype=str)
-frontends = ['16_pcenMel128', '16_logMel128', '16_logSTFT', '16_Mel128', '8_pcenMel64', '32_pcenMel128']
+frontends = ['16_logMel128', '16_logSTFT', '16_Mel128', '16_pcenMel128', '8_pcenMel64', '32_pcenMel128', '64_pcenMel128']
 plt.figure()
 for specie in species:
     df = pd.read_csv(f'{specie}/{specie}.csv')
@@ -30,3 +30,4 @@ plt.grid()
 plt.tight_layout()
 plt.legend()
 plt.savefig('NMIs_hdbscan.pdf')
+plt.close()

print_annot.py

@@ -8,7 +8,7 @@ import torch
 parser = argparse.ArgumentParser()
 parser.add_argument("specie", type=str)
 parser.add_argument("-frontend", type=str, default='logMel')
-parser.add_argument("-nMel", type=int, default=64)
+parser.add_argument("-nMel", type=int, default=128)
 args = parser.parse_args()
 
 meta = models.meta[args.specie]

run_baseline.py

@@ -12,8 +12,8 @@ args = parser.parse_args()
 
 df = pd.read_csv(f'{args.specie}/{args.specie}.csv')
 
-def norm(arr):
-    return (arr - np.mean(arr) ) / np.std(arr)
+norm = lambda arr: (arr - np.mean(arr) ) / np.std(arr)
+
 meta = models.meta[args.specie]
 
 feats = ['fund', 'cvfund', 'maxfund', 'minfund', 'meansal', 'meanspect', 'stdspect', 'skewspect',\
@@ -32,7 +32,7 @@ def process(idx):
     if fs != meta['sr']:
         sig = resample(sig, int(len(sig)/fs*meta['sr']))
     sound = BioSound(soundWave=norm(sig), fs=fs)
-    sound.spectroCalc(max_freq=meta['sr']//2)
+    sound.spectroCalc(max_freq=meta['sr']//2, spec_sample_rate=128//meta['sampleDur'])
     sound.rms = sound.sound.std() 
     sound.ampenv(cutoff_freq = 20, amp_sample_rate = 1000)
     sound.spectrum(f_high=meta['sr']//2 - 1)
@@ -43,9 +43,9 @@ def process(idx):
     
     return [sound.__dict__[f] for f in feats]
 
-res = p_tqdm.p_map(process, df.index[:10])
+res = p_tqdm.p_map(process, df.index[:100], num_cpus=16)
 
-for i, mr in zip(df.index[:10], res):
+for i, mr in zip(df.index[:100], res):
     for f, r in zip(feats, mr):
         df.loc[i, f] = r
 

run_vggish.py

+from sklearn import metrics
+import matplotlib.pyplot as plt
+import umap, hdbscan
+from tqdm import tqdm
+import argparse, os
+import models, utils as u
+import pandas as pd, numpy as np, torch
+torch.multiprocessing.set_sharing_strategy('file_system')
+
+
+parser = argparse.ArgumentParser()
+parser.add_argument("specie", type=str)
+args = parser.parse_args()
+
+df = pd.read_csv(f'{args.specie}/{args.specie}.csv')
+
+meta = models.meta[args.specie]
+
+if not os.path.isfile(f'{args.specie}/encodings_vggish.npy'):
+    gpu = torch.device('cuda')
+    frontend = models.frontend['logMel_vggish'](meta['sr'], meta['nfft'], meta['sampleDur'], 64)
+    vggish = torch.hub.load('harritaylor/torchvggish', 'vggish')
+    # vggish.preprocess = False
+    vggish.postprocess = False
+    model = torch.nn.Sequential(frontend, vggish).to(gpu)
+    model.eval()
+    loader = torch.utils.data.DataLoader(u.Dataset(df, f'{args.specie}/audio/', 16000, 1), batch_size=1, shuffle=True, num_workers=8, collate_fn=u.collate_fn)
+    with torch.no_grad():
+        encodings, idxs = [], []
+        for x, idx in tqdm(loader, desc='test '+args.specie, leave=False):
+            # encoding = model(x.to(gpu))
+            encoding = vggish(x.numpy().squeeze(0), fs=16000)
+            idxs.extend(idx)
+            encodings.extend(encoding.cpu().detach())
+
+    idxs, encodings = np.array(idxs), np.stack(encodings)
+    X = umap.UMAP(n_jobs=-1).fit_transform(encodings)
+    np.save(f'{args.specie}/encodings_vggish.npy', {'idxs':idxs, 'encodings':encodings, 'umap':X})
+else:
+    dic = np.load(f'{args.specie}/encodings_vggish.npy', allow_pickle=True).item()
+    idxs, encodings, X = dic['idxs'], dic['encodings'], dic['umap']
+
+clusters = hdbscan.HDBSCAN(min_cluster_size=50, min_samples=5, cluster_selection_epsilon=0.05, core_dist_n_jobs=-1, cluster_selection_method='leaf').fit_predict(X)
+df.loc[idxs, 'cluster'] = clusters.astype(int)
+mask = ~df.loc[idxs].label.isna()
+
+#print('Found clusters : \n', pd.Series(clusters).value_counts())
+
+plt.figure(figsize=(20, 10))
+plt.scatter(X[clusters==-1,0], X[clusters==-1,1], s=2, alpha=.2, color='Grey')
+plt.scatter(X[clusters!=-1,0], X[clusters!=-1,1], s=2, c=clusters[clusters!=-1], cmap='tab20')
+plt.tight_layout()
+plt.savefig(f'{args.specie}/vggish_projection_clusters.png')
+
+plt.figure(figsize=(20, 10))
+plt.scatter(X[~mask,0], X[~mask,1], s=2, alpha=.2, color='Grey')
+for l, grp in df.groupby('label'):
+    plt.scatter(X[df.loc[idxs].label==l, 0], X[df.loc[idxs].label==l, 1], s=4, label=l)
+plt.legend()
+plt.tight_layout()
+plt.savefig(f'{args.specie}/vggish_projection_labels.png')
+
+
+clusters, labels = clusters[mask], df.loc[idxs[mask]].label
+print('Silhouette', metrics.silhouette_score(encodings[mask], clusters))
+print('NMI', metrics.normalized_mutual_info_score(labels, clusters))
+print('Homogeneity', metrics.homogeneity_score(labels, clusters))
+print('Completeness', metrics.completeness_score(labels, clusters))
+print('V-Measure', metrics.v_measure_score(labels, clusters))
+
+labelled = df[~df.label.isna()]
+for l, grp in labelled.groupby('label'):
+    best = (grp.groupby('cluster').fn.count() / labelled.groupby('cluster').fn.count()).idxmax()
+    print(f'Best precision for {l} is for cluster {best} with {(df.cluster==best).sum()} points, \
+with precision {((labelled.cluster==best)&(labelled.label==l)).sum()/(labelled.cluster==best).sum():.2f} and recall {((labelled.cluster==best)&(labelled.label==l)).sum()/(labelled.label==l).sum():.2f}')

test_AE.py

@@ -7,6 +7,7 @@ from sklearn import metrics
 import umap, hdbscan
 torch.multiprocessing.set_sharing_strategy('file_system')
 
+
 parser = argparse.ArgumentParser()
 parser.add_argument("specie", type=str)
 parser.add_argument("-bottleneck", type=int, default=16)
@@ -21,8 +22,6 @@ df = pd.read_csv(f'{args.specie}/{args.specie}.csv')
 print(f'Tests for model {modelname}')
 print(f'{len(df)} available vocs')
 
-loader = torch.utils.data.DataLoader(u.Dataset(df, f'{args.specie}/audio/', meta['sr'], meta['sampleDur']), batch_size=64, shuffle=True, num_workers=8, collate_fn=u.collate_fn)
-
 if os.path.isfile(f'{args.specie}/encodings_{modelname[:-4]}npy'):
     dic = np.load(f'{args.specie}/encodings_{modelname[:-4]}npy', allow_pickle=True).item()
     idxs, encodings, X = dic['idxs'], dic['encodings'], dic['umap']
@@ -34,6 +33,7 @@ else:
     model = torch.nn.Sequential(frontend, encoder, decoder).to(gpu)
     model.load_state_dict(torch.load(f'{args.specie}/{modelname}'))
     model.eval()
+    loader = torch.utils.data.DataLoader(u.Dataset(df, f'{args.specie}/audio/', meta['sr'], meta['sampleDur']), batch_size=64, shuffle=True, num_workers=8, collate_fn=u.collate_fn)
     with torch.no_grad():
         encodings, idxs = [], []
         for x, idx in tqdm(loader, desc='test '+args.specie, leave=False):
@@ -45,7 +45,7 @@ else:
     X = umap.UMAP(n_jobs=-1).fit_transform(encodings)
     np.save(f'{args.specie}/encodings_{modelname[:-4]}npy', {'idxs':idxs, 'encodings':encodings, 'umap':X})
 
-clusters = hdbscan.HDBSCAN(min_cluster_size=10, min_samples=5, cluster_selection_epsilon=0.0, core_dist_n_jobs=-1, cluster_selection_method='eom').fit_predict(X)
+clusters = hdbscan.HDBSCAN(min_cluster_size=50, min_samples=5, cluster_selection_epsilon=0.05, core_dist_n_jobs=-1, cluster_selection_method='leaf').fit_predict(X)
 df.loc[idxs, 'cluster'] = clusters.astype(int)
 mask = ~df.loc[idxs].label.isna()
 
@@ -74,7 +74,15 @@ print('Completeness', metrics.completeness_score(labels, clusters))
 print('V-Measure', metrics.v_measure_score(labels, clusters))
 
 labelled = df[~df.label.isna()]
+goodClusters = []
 for l, grp in labelled.groupby('label'):
-    best = (grp.groupby('cluster').fn.count() / labelled.groupby('cluster').fn.count()).idxmax()
+    precisions = grp.groupby('cluster').fn.count() / labelled.groupby('cluster').fn.count()
+    best = precisions.idxmax()
+    goodClusters.extend(precisions[precisions > 0.9].index)
     print(f'Best precision for {l} is for cluster {best} with {(df.cluster==best).sum()} points, \
-with precision {((labelled.cluster==best)&(labelled.label==l)).sum()/(labelled.cluster==best).sum():.2f} and recall {((labelled.cluster==best)&(labelled.label==l)).sum()/(labelled.label==l).sum():.2f}')
+    with precision {((labelled.cluster==best)&(labelled.label==l)).sum()/(labelled.cluster==best).sum():.2f}\
+     and recall {((labelled.cluster==best)&(labelled.label==l)).sum()/(labelled.label==l).sum():.2f}')
+
+
+print(f'{len(goodClusters)} clusters would sort {df.cluster.isin(goodClusters).sum()/len(df)*100:.0f}% of samples')
+print(f'{len(goodClusters)/df.label.nunique():.1f} cluster per label in avg)')
\ No newline at end of file

train_AE.py

@@ -67,8 +67,10 @@ for epoch in range(100_000//len(loader)):
 
         if len(loss) > 2000 and np.median(loss[-2000:-1000]) < np.median(loss[-1000:]):
             print('Early stop')
+            torch.save(model.state_dict(), f'{args.specie}/{modelname}')
             exit()
-
+        step += 1
+        continue
         # TEST ROUTINE
         if step % 500 == 0:
             # Plot reconstructions