Accessible AI Smart Assistant

Inspiration

The inspiration for this project was to create a cheap-to-implement AI chatbot that was as realistic as possible. This code would act as the framework for a sort of smart assistant, like Alexa or Google Home, but with much more functionality and an actual personality. Users could tailor their assistant's personality (to an extent). Some additional code and prompts would be required to teach the AI commands that the Alexa and Home have, but the beauty of the AI chatbot is that it could be taught new commands by the user. The ability for users to speak and simply hear their chatbot makes them more accessible to users with limited mobility, limited vision, and other impairments.

What it does

This project uses google speech recognition to listen to a user and convert their speech to text. Then, it inputs this text into the GPT-3 language processing engine, with the goal of simulating a conversation between the user and the GPT-3 engine. It will return text, which is then turned into speech and played to the user.

How we built it

This project was built using python and OpenAI. Before this hackathon, I had access to the gpt-3 engine, but no code was written.

Challenges we ran into

Attempted to use Whisper from OpenAI, but certain requirements for Whisper made using google speech recognition more desirable.

What we learned

1. How to tailor responses from GPT-3 AI
2. Sending text to GPT-3 receiving a response
3. Using PyAudio framework to implement speech recognition
4. Outputting audio using python framework

Built With

• openai
• pyaudio
• python