Fowl Escapades
Helping Africa
2000 Zindi Points
Completed (over 6 years ago)
Classification
199 joined
28 active
Start
16 Jan 20
Close
19 Apr 20
Reveal
20 Apr 20
 scale(.21174)'%3e%3cpath d='m52.487 177.086 13.018 11.852 11.762-11.762 11.896 12.266 12.846-11.358 11.827 12.43 13.339-11.013 10.897 12.728 13.27-10.145 13.947 12.319'/%3e%3ccircle cx='89.425' cy='154.584' r='5.167'/%3e%3ccircle cx='111.237' cy='156.047' r='5.167'/%3e%3ccircle cx='30.05' cy='16.5' r='4'/%3e%3c/g%3e%3c/svg%3e)
Getting Started
Notebooks · 16 Jan 2020, 15:59 · 2
I threw together a quick starter notebook: https://colab.research.google.com/drive/1SdrwQtT16FAGmfssJi5riMjDOQU7IvWE
It uses fastai, treating this as essentially an image classification task. As you can imagine, this isn't the optimal solution! That said, it's possible to get some pretty good performance with this method.
It scores ~2.7. Accuracy is only ~32%, but it gets the right bird in the top 5 guesses >60% of the time.
Hoping to see some different techniques springing up (please share what you're trying!!) but this at least will show an example path of loading the data, extracting features (spectrograms in this case), training a model and making predictions.
Good luck all :)
Great notebook.
The "Get Creative" part at the end got me pretty excited. I'm going to try some crazy ideas.
Sweet, thanks for the starter code! It's pretty creative to use the spectrogram and to turn this into an image classification problem, but as you say it is probably not optimal.
For those who want to explore the topics of audio and signal processing further, this is a handy guide that goes from noob to pro pretty quickly. To clarify, the noob parts are suitable for anyone who is interested in how audio is processed digitally. This is general knowledge. The pro parts of this post involve more intense Python, so if you don't want to use Python, then you may safely skip that part.
https://www.analyticsvidhya.com/blog/2017/08/audio-voice-processing-deep-learning/