Inspiration
A problem space in speech where people with speech impairments are unable to communicate effectively with other people.
What it does
We built an AI voice assistant which autocompletes and finished sentences if the input sentence contains elements such as stuttered words or the sentence is incomplete.
How we built it
We used OpenAI models whisper for getting audio transcriptions from user voice input and gpt 4 for generating responses. on the client side we use gradio for the UI.
Challenges we ran into
Our idea originally incorporated a vector store (FAISS) and OpenAI embeddings to augment gpt 4 responses by finding similar sentences. Mapping impaired speech to correct speech was a challenge as dataset was not fully complete.
Accomplishments that we're proud of
We produced a working product where the AI assistant is able to complete and formulate correct sentences for the user.
Built With
- openai
- python
Log in or sign up for Devpost to join the conversation.