​Project Link

Jeju Island is a popular tourist destination in Korea famous for its pristine environment. It is also a beloved place for birdwatchers, and also the place where my grandparents lived.

This is a birdsong classifier that analyzes your voice and tells you which type of Jeju bird you are. Try whistling into the mic and mimic some bird sounds! ​​

Bird species

There are currently 13 species available:

• Japanese Night Heron (Gorsachius goisagi)
• Red-necked Stint (Calidris ruficollis)
• Japanese Murrelet (Synthliboramphus wumizusume)
• Large-billed Crow (Corvus macrorhynchos)
• Varied Tit (Sittiparus varius varius)
• Japanese Tit (Parus minor minor)
• Brown-eared Bulbul (Hypsipetes amaurotis amaurotis)
• Japanese Bush Warbler (Horornis diphone cantans)
• Pale Thrush (Turdus pallidus)
• Dusky Thrush (Turdus eunomus)
• Grey-capped Greenfinch (Chloris sinica)
• Meadow Bunting (Emberiza cioides),
• Yellow-throated Bunting (Emberiza elegans)

Dataset

Xeno-canto is a website where you can find birdsongs from different parts of the world. Using this website I was able to download birdsongs from 13 species of birds inhabiting in Jeju Island. I omitted some species because some of the uploaded audio files were too short for training.

Training a classifier

Using Teachable Machine, I was able to train an audio classifier that classifies bird sounds from audio input.

Since Teachable Machine does not support uploading audio files directly, I extracted the audio samples by playing the audio files of 13 birdsongs manually.

To train the model, I also needed some background noise. I manually recorded some background noises from the birdsong audio files I downloaded from the website.

You can check out the trained model in this link.