Acoustic Keystroke Classifier

Inspiration

To learn more about machine learning and how it can be used to classify audio.

What it does

Tries to classify keys based off of the keystroke sound.

How we built it

We implemented a CNN with 3 convolutional layers.

Challenges we ran into

Processing the data into matrices we could feed to the model was a challenge. Also, recording the data for training as well as predicting was a challenge because we had to find a way to minimize the noise that triggered the code even when no key was pressed. Other challenges included implementing the web application to demo the model.

Accomplishments that we're proud of

We achieved ~50% accuracy while training, on a validation data set, which was our best result.

What we learned

We learned about convolutional neural networks and how they can be used to classify audio. We learned how to use flask to integrate the models functionality into an application.

What's next for Acoustic Keystroke Classifier

Adding a recurrent neural network which would follow the CNN, to account for statistical data from the audio recording such as time between key presses, and which would account for personal typing styles and patterns.

Built With

• css
• flask
• html
• javascript
• numpy
• python
• pytorch