I2T-ASL (Text to ASL)

• 

  1

• 

  4

• 

  9

Inspiration

AI text-to-image generators are contributing to the notion of artistic practice. This team of artists and computer scientists on this team acknowledge the d/Deaf community, for this project seeks to reduce inequality and increase the well-being for the d/Deaf community.

What it does

This website consists of an image-to-text AI program that currently interprets the numbers 1-9 in American Sign Language. As an outlet for the d/Deaf community, we seek to show how their signals can be expressed by an AI program through text and art.

How we built it

This website was built with HTML and CSS and integrated with JavaScript and TensorFlow. The images were created with Stable Diffusion, through Hypergator.

Challenges we ran into

Using Google's Teachable Machine to train our own image model was the most difficult task. Determining the best hyperparameters at which the model worked most effectively was important.

Accomplishments that we're proud of

Our most proud accomplishment was training our own image model and making it work with TensorFlow.

What we learned

We learned more about making an image classifier model, and we learned how to integrate it into a web app. We had to make predictions from the model and figure out how to make it work for our purposes.

What's next for I2T-ASL

For the future of I2T-ASL, we’d make a better image model with the alphabet, more characters, and signed phrases. We want to have a more streamlined way of generating artwork from an image-to-text interpretation using an API.

Built With

• css
• html
• javascript
• stable-diffusion
• tensor-flow