Sound detection of elephants using machine learning
Budget: $250 – $750 USD
First step:
We need an AI that is able to detect elephant rumbles from jungle recordings.
Additionally:
The AI should show it's predictions (time-stamps from a sound file) and we should be able to tell the AI if it is true or not. So we can train the AI further with new recordings.
A GUI would be nice to have. But that depends if we can afford it :)
What we already have:
Our dataset exists of many hours jungle recordings with around 1000 labeled elephant sounds in it. That should be ready to feed the AI.
More info about the elephant sounds:
Elephants make different sounds but rumbles have a very low frequency (around 20-50Hz) so they can be still picked up from a microphone even after 2Km in good conditions. The AI should be able to detect also these further rumbles.
How that all looks like in a spectrogram you can see here in this video:
https://youtu.be/XB6pt4CHhA4?t=40
With your input you will be a huge help to reduce human-elephant-conflict, which is the number one reason for elephant deaths.
We are a team of around 10 people. I am volunteering since one year for this project (The hardware guy) and hope you can help us to bring this project to reality. As a non profit organization our budget is limited. Hope you can consider that as well :)
At the end (Not part of your job for now):
The trained model should run on a solar powered microcontroller (ESP32) as an elephant detector while it is constantly listening in the jungle with a connected microphone. These detectors will be able to spot elephants before they reach a farm or village and warn rangers in that area who can deal with that situation. This will prevent human-elephant-conflict.
As far as i know only Tensorflow (With Tesnorflow Lite) or PyTorch (with TinyML)are able to run on a ESP32. Which should be considered.
If you have any further questions. Feel free to contact me
We need an AI that is able to detect elephant rumbles from jungle recordings.
Additionally:
The AI should show it's predictions (time-stamps from a sound file) and we should be able to tell the AI if it is true or not. So we can train the AI further with new recordings.
A GUI would be nice to have. But that depends if we can afford it :)
What we already have:
Our dataset exists of many hours jungle recordings with around 1000 labeled elephant sounds in it. That should be ready to feed the AI.
More info about the elephant sounds:
Elephants make different sounds but rumbles have a very low frequency (around 20-50Hz) so they can be still picked up from a microphone even after 2Km in good conditions. The AI should be able to detect also these further rumbles.
How that all looks like in a spectrogram you can see here in this video:
https://youtu.be/XB6pt4CHhA4?t=40
With your input you will be a huge help to reduce human-elephant-conflict, which is the number one reason for elephant deaths.
We are a team of around 10 people. I am volunteering since one year for this project (The hardware guy) and hope you can help us to bring this project to reality. As a non profit organization our budget is limited. Hope you can consider that as well :)
At the end (Not part of your job for now):
The trained model should run on a solar powered microcontroller (ESP32) as an elephant detector while it is constantly listening in the jungle with a connected microphone. These detectors will be able to spot elephants before they reach a farm or village and warn rangers in that area who can deal with that situation. This will prevent human-elephant-conflict.
As far as i know only Tensorflow (With Tesnorflow Lite) or PyTorch (with TinyML)are able to run on a ESP32. Which should be considered.
If you have any further questions. Feel free to contact me