Autodidact

💡 Inspiration

When I was younger, I used to stutter quite a bit, and generally, I was not great at communicating.. It was a big problem, but funnily enough, the solution was quite simple. I saw a video taken of me, giving a speech to the entire class, and I marveled at how poor my communication skills were, and in a matter of weeks, I had gotten rid of my stutter and posture. It is

🤷‍♂️ What it does

It is a simple program that records your voice and plays it back to you so you can hear what you sound like. Pull up a script and speak, and you get instant feedback on how you sound. Also, you have a text-to-speech function that also indicates how clearly the ML model understood you.

🧱 How I built it

I built this using python in google colab. I used HTML for the voice recorder and used facebook's pre-trained NLP model for text-to-speech.

🧩 Challenges we ran into

In general, this was not a great week for a hackathon. I had finals to study for, end-of-quarter projects, and a lot of clubs wrapping up. It was difficult to portion my time and I got things done pretty last minute. Another huge challenge was having to remove features because they would not work. Initially, Autodidact was supposed to take a video recording of your voice and face as you practice speaking/presenting, and it would use ML to classify your facial expressions and tone while talking to rate different factors of your dictation and presentation, such as confidence in speech and look, gaps/filler words, etc... As it turns out, this was a gargantuan task and so I scraped together what I could. I also wanted to build a streamlit GUI for my project, but my pip packages refused to work halfway through so I moved to colab and salvaged my code.

🎉 Accomplishments that we're proud of

I am proud that I was able to get something on the table, regardless of the obstacles with my code and irl. I also am proud that this is my first machine learning hack, I learned about a lot of technologies and while I couldn't implement them all, this knowledge will definitely be useful in the future.

🏫 What we learned

I learned about natural language processing and neural networks, and how they work and can be trained to classify and predict nearly anything given enough data. I also learned a little about psychology and suggestion therapy, and was quite intrigued. Overall, I learned many new things this hackathon, not just about technologies but real world lessons and concepts.

⏭ What's next for Autodidact

In a later hackathon, I hope to build on Autodidact and make it the monster I originally intended it to be. First, I need to get the models working for facial features and speech classification while communicating, so I have to find models for that. Second, I want to build a nice web application that makes this easy and intuitive to use, and ideally host it publicly so anyone can use it to improve themselves. Autodidact refers to someone who is self-taught, and with Autodidact I can help others undergo the transformation I did, and be confident in their speaking abilities.

Built With

• ml
• python