Umazing! Elephant Rumbles

Inspiration

For this project, we wanted to focus on elephant rumbles, as they are the hardest to hear compared to their loud trumpet counterparts, and found that isolating them would prove more useful in the field. We also decided to eliminate specifically airplane noise, as it sounds similar to elephant rumbles.

To do this, we converted each .wav file into a spectrogram and, using a trained UNet model, isolated the elephant rumbles by analyzing the specific frequencies they typically fall between. Then the program returns the isolated.wav file for the listener to hear.

Challenges

Some challenges we faced included determining what architecture to use for the machine learning process and how to isolate the waves in the first place, given that the human brain processes audio very differently from a computer.

A more specific challenge came from our earlier planning stages. At first, we decided to host the website on GitHub Pages, but later we wanted to add a Gemini API for some elephant facts. Because GitHub Pages can only host static websites, we were unable to add API calls and instead replaced the call with a list of randomly selected facts to display.

Takeaways

As a group, we learned how to analyze audio for isolation, including by examining waveforms and spectrograms. We also learned about machine learning models and how to analyze loss diagrams. Finally, on a broader note, we learned about conservation efforts for elephants and the importance of learning about the world around us, and how biology and technology rely on one another.

Built With

• flask
• github
• html
• python
• pytorch