Table 1.
Data set | Trill | Long whistle | Short whistle | Tsak | Zip | |||||
---|---|---|---|---|---|---|---|---|---|---|
Ncalls | NInd/groups | Ncalls | NInd/groups | Ncalls | NInd/groups | Ncalls | NInd/groups | Ncalls | NInd/groups | |
Training data set | ||||||||||
Detection (2123 calls) | 326 | 12 | 159 | 4 | 1109 | 6 | 519 | 5 | 10 | 1a |
Classification and clustering (2257 calls) | 302 | 52 | 186 | 24 | 1158 | 46 | 541 | 34 | 70 | 20 |
Standardized data set | ||||||||||
Good-quality (450 calls) | 50 | 42 | 50 | 22 | 150 | 40 | 150 | 29 | 50 | 19 |
Clipped (80 calls) | 10 | 9 | 10 | 8 | 30 | 9 | 30 | 10 | – | – |
Low-amplitude (80 calls) | 10 | 7 | 10 | 10 | 30 | 10 | 30 | 10 | – | – |
Overlaid (80 calls) | 10 | 8 | 10 | 8 | 30 | 8 | 30 | 9 | – | – |
Original experimental data set | ||||||||||
M. murinus (3040 calls) | 76 | 7 | 157 | 5 | 2582 | 8 | 195 | 6 | 30 | 7 |
M. lehilahytsara (1115 calls) | 48 | 5 | 526 | 5 | 240 | 1 | 193 | 3 | 108 | 4 |
aZips occur rarely in a sufficient signal-to-noise ratio. However, because they are highly stereotyped, this low number of Zips turned out to be sufficient to train the detector.