Inspiration
What is one of the biggest motivators at hackathons? The thrill of competition. What if you could bring that same excitement into your daily life, using your speech to challenge your friends? Our app gamifies everyday conversations and interactions, letting you and your friends compete to see who projects the most positive or negative aura.
What it does
- Captures live audio input and converts it into text using React’s speech recognition library.
- Analyzes the transcribed text by running it through Cohere’s semantic similarity model by encoding both input data and our dataset into vector embeddings
- Uses cosine similarity to compare the input with synthetic data, generated using ChatGPT, in order to evaluate whether the speech conveys a positive or negative aura based.
Challenges we ran into
- Integrating real-time speech recognition with accurate transcription, especially when dealing with diverse speech patterns and accents.
- Acquiring a continuous audio input which can then be passed along for efficient transcription.
- Configuring Cohere’s API to work seamlessly with a large dataset and ensure fast, accurate sentiment analysis.
- Getting accurate data on words/actions that constitute "positive aura" and "negative aura".
Accomplishments that we're proud of
- Cohere Embeddings for Sentiment Analysis: Integrating Cohere’s powerful semantic embeddings was another significant milestone. We used their embeddings to analyze and determine the sentiment of transcribed text, mapping speech patterns to either positive or negative aura. We’re proud of this implementation because it brought depth to the app.
What's next for Traura
- Turning this prototype into a full-fledged web app that users can access anywhere, including the full implementation of the leaderboard functionality to foster that friendly thrill of competition.
Log in or sign up for Devpost to join the conversation.