18 new birds and other sounds now recognised by the bird monitor

We've just rolled out a couple of changes to our bird sound processing that now recognises 18 new birds and other sounds. We also automatically create boxes around the sounds that we are identifying.

We are actively working on some other improvements.

New birds

Thanks to those that have been listening to recordings and tagging birds we can now automatically recognise:

  1. bellbird
  2. black noddy
  3. blackbird
  4. crimson rosella
  5. fantail
  6. frog
  7. grey warbler
  8. house sparrow
  9. insect
  10. noise:this is a generic sound that isn't a bird or some other sound that we're trying to recognise.
  11. norfolk gerygone
  12. norfolk robin
  13. rifleman
  14. rooster
  15. silvereye
  16. sooty tern
  17. sparrow
  18. white tern

This is in addition to the birds/sounds we could already recognise

  1. human,
  2. kiwi,
  3. morepork,
  4. whistler 
  5. generic bird calls.

Special thanks to those that have been listening to the audio recordings. With a little bit of trans Tasman rivalry, the Norfolk Island taggers have done a fantastic job, tagging almost 20,000 calls. There have also been many calls tagged by the Sumner Bays group and the Extended Wildside on Banks Peninsula. Thank you to everyone who has helped here. Thanks also to Giampaolo (GP) who has taken all of this data and created these models.

This release is a little bit of a test. We are not sure how well it will work because for some of these birds we don't have a lot of data to train and test. This means you will expect to see classification mistakes. Any mistakes that you do see, please just correct the classification and that will help us further improve the models.

Automatic boxes

A new algorithm automatically draws boxes around each sound that is identified. This may make the tagging process easier because you don't have to draw boxes. However it can mean there are a lot of boxes on some recordings, like the one below.

Spectrogram with each sound automatically identified

Future Improvements

We are actively working on future improvements.

Automatically delete human voices

We are about to start testing an optional feature that will automatically delete any recordings that have human voices. This is to protect the privacy of those who might be near the bird recorders.

Interface improvements

We have had a number of suggestions from our customers and within our team to improve the bird tagging interface. We are about to start working on these improvements.

New camera includes a microphone

Our new thermal camera (which is cheaper and better than our old one) includes a microphone. We intend to add the ability to use the camera as an audio bird monitor at some stage down the track. If you buy a camera you will hopefully have a bird monitor as well. This camera is available for pre-order now

New bird monitor

You may have noticed that we have stopped selling our existing bird monitor. The reason for this is that we have been unable to source cheap android phones. The cheapest phones we can find now would push the price of the bird monitors up considerably.

For many years The Cacophony Project have talked about developing the bird monitor hardware ourselves but we have never been able secure funding. We have been working on this in our spare time and hope to finish this after we have completed the hardware design for our new thermal camera. We hope it will be cheaper and more robust that the existing bird monitors. Watch this space and please let us know if you're interested.

Processing old recordings

Once we are confident that our model is producing reasonable results we are planning on processing old bird recordings. This will give you data on all your historical recordings. 

Confusion matrices

These charts show the accuracy of our model against the test data for the different birds. For each bird this shows you the number of times that our model correctly identified a bird against the test data and the number of times it got it wrong. It also shows you the number of times we incorrectly identified a sound as that bird.

For, example the chart below shows that 70% of bellbirds in the test set were correctly identified as bellbirds and 30% weren't. It also shows that none of the other sounds were incorrectly identified as a bellbird.

Bellbird confusion matrixBird confusion matrixblack noddy confusion matrixBlackbird confusion matrixFantail confusion matrixFrog confusion matrixHuman confusion matrixInsect confusion matrixMorepork confusion matrixNoise confusion matrixNorfolk GreygoneRooster confusion matrixSparrow confusion matrix

Leave a comment

Please note, comments must be approved before they are published