Birds recognised expanded to over 60

The DOC AI Cam includes a microphone and can optionally be configured to make audio recordings.

We have just released our latest bird recognition model that greatly expands the number of birds that we recognise to over 60. More accurately, it's over 60 sounds because this includes insect, frog, weta, human and noise.

This expanded list has been trained using tagged bird recordings supplied by DOC. Thank you very much to them for sharing this with us. When we tested the model against our test set it did a pretty good job at identifying the birds accurately, or if it was not able to do that it would identify them as "don't know".  This should reduce the number of misclassifications.

Our algorithm is still using the data from eBird to restrict, by region which birds will be identified. Thank you to them too for letting us use this data.

The real test is how it goes in the wild (literally). Please let us know how well the new algorithm is doing.

Here's the complete list. This shows how the model performs against our test set. You can see that while the  accuracy isn't that good at 42%, it will still label it as bird 31% which is accurate, but not specific. 6% of the time it will say unsure. It get's it wrong about 21% of the time. You can see in the table below what the most common mis-classified bird is. You can also see the number of samples we have for each bird in our test set (note this is different from the samples we use for training).  These numbers should improve as we get more accuracy.

label correct % misclassified label misclassified % unsure % bird % total samples
Average 42.3% 12.9% 6.1% 30.8%
bellbird 19.9% noise 9.3% 7.7% 53.5% 2014
black noddy 87.8% white tern 2.0% 2.0% 8.2% 49
black-swan 0.0% human 50.0% 50.0% 0.0% 2
blackbird 24.9% noise 11.3% 8.8% 43.3% 799
brown-creeper 32.6% noise 6.5% 6.5% 43.5% 46
california quail 0.0% fantail 20.0% 20.0% 20.0% 5
chaffinch 33.0% noise 11.3% 5.2% 38.7% 442
crimson rosella 37.1% white tern 4.8% 0.0% 50.0% 62
duck 25.0% noise 25.0% 0.0% 50.0% 4
dunnock 25.9% chaffinch 7.4% 14.8% 25.9% 27
fantail 45.7% noise 9.9% 4.9% 31.9% 698
frog 0.0% noise 94.1% 5.9% 0.0% 17
goose 28.6% noise 28.6% 28.6% 14.3% 7
greenfinch 11.1% blackbird 5.6% 2.8% 55.6% 36
grey warbler 46.9% noise 10.3% 6.7% 30.3% 1022
indian-myna 0.0% None 0.0% 20.0% 80.0% 5
insect 81.6% noise 13.1% 2.2% 1.5% 267
kea 60.6% noise 7.9% 4.7% 17.3% 127
kereru 0.0% noise 28.6% 0.0% 42.9% 7
kiwi 42.7% noise 18.5% 5.6% 7.9% 178
long-tailed-cuckoo 61.1% noise 18.7% 3.3% 10.7% 337
magpie 33.3% morepork 16.7% 8.3% 16.7% 12
marsh-crake 0.0% noise 100.0% 0.0% 0.0% 2
mohua 20.0% noise 40.0% 0.0% 0.0% 5
morepork 65.2% noise 21.2% 8.6% 4.3% 2722
new-zealand-falcon 8.3% chaffinch 8.3% 33.3% 25.0% 12
new-zealand-fernbird 0.0% noise 50.0% 0.0% 50.0% 2
new-zealand-kaka 49.8% noise 8.0% 6.7% 27.1% 538
new-zealand-kingfisher 26.3% noise 10.5% 5.3% 34.2% 38
new-zealand-pipit 22.2% noise 44.4% 22.2% 11.1% 9
norfolk gerygone 39.0% blackbird 3.5% 0.7% 44.0% 141
norfolk robin 52.2% white tern 4.3% 0.0% 36.2% 69
north-island-robin 16.3% noise 6.7% 3.8% 54.8% 104
oystercatcher 25.0% None 0.0% 0.0% 75.0% 4
paradise-shelduck 40.0% noise 20.0% 0.0% 40.0% 5
parakeet 27.0% noise 11.5% 5.7% 47.5% 122
pukeko 64.4% noise 8.9% 2.2% 13.3% 45
red-tailed tropicbird 12.5% black noddy 25.0% 12.5% 50.0% 8
redpoll 25.0% chaffinch 5.0% 12.5% 42.5% 40
rifleman 31.5% noise 24.2% 5.0% 31.0% 400
rooster 65.7% noise 2.9% 0.0% 23.5% 102
shining-cuckoo 60.0% noise 7.5% 0.0% 22.5% 40
silvereye 43.4% noise 7.3% 4.6% 38.0% 1564
skylark 45.8% noise 12.5% 4.2% 25.0% 24
slender-billed white-eye 16.5% silvereye 17.7% 1.3% 53.2% 79
song thrush 22.1% noise 11.8% 5.9% 46.3% 136
sooty tern 42.9% black noddy 17.1% 0.0% 25.7% 35
south-island-robin 19.5% noise 11.9% 6.8% 51.7% 118
southern-black-backed-gull 33.3% noise 33.3% 0.0% 16.7% 6
sparrow 36.4% silvereye 31.8% 0.0% 27.3% 22
spur-winged-plover 22.2% noise 44.4% 0.0% 11.1% 9
starling 25.0% song thrush 25.0% 12.5% 0.0% 8
tomtit 16.9% noise 13.6% 6.6% 48.8% 1620
tree-weta 58.8% noise 25.1% 1.4% 0.0% 211
tui 13.5% noise 6.0% 7.9% 57.2% 554
weka 60.4% noise 16.7% 4.2% 11.1% 144
whistler 62.9% fantail 2.1% 0.6% 28.2% 869
white tern 73.1% black noddy 3.2% 0.0% 15.1% 93
whitehead 24.4% noise 9.8% 2.4% 46.3% 41
yellow-crowned-parakeet 23.8% rifleman 9.5% 0.0% 47.6% 21
yellowhammer 22.2% noise 18.5% 11.1% 37.0% 27
bird 41.0% noise 16.9% 15.7% 41.0% 83
human 49.1% bird 14.0% 22.8% 14.0% 57
noise 66.9% insect 23.2% 4.1% 1.7% 708

1 comment

  • Fabulous work – hope to see kōkako on the next list!

    Inger Perkins

Leave a comment

Please note, comments must be approved before they are published