Eloquence

• 

  Home page

• 

• 

• 

  Analysis pt1

• 

  Analysis pt2

Inspiration

Our project was completely developed originally. If we were to cite an inspiration, it would be Grammarly but instead of analyzing text, we analyze your speech.

We were inspired by our own lives at school and the lives of our family and friends. A tool like this would have helped us greatly throughout our years at school, and family members who have done business presentations have also let us know how a tool like this would be helpful for them.

What it does

Eloquence analyzes your speech patterns and provides you with helpful stats, feedback, and visuals. Simply upload an audio file of you speaking, and Eloquence will do the rest.

How we built it

We built Eloquence using svelte-kit + SkeletonUI + threlte for the front end and python for the back end. Specifically, we used Blacksheep to run our backend server, and svelte-kit to run our frontend website. We used the Distil-Whisper: distil-large-v2 Hugging Face model to convert the speech provided into text, while maintaining the filler words we wanted to detect. We also used the OpenAI API to provide detailed feedback to the user about what they could improve specifically within their speech, presentation, etc.

Challenges we ran into

Many speech-to-text models we found could not detect filler words so we had to use a machine-learning model to detect everything. Additionally, we ran into problems when prompt engineering the OpenAI responses, as well as optimizing both the OpenAI and HuggingFace models.

Accomplishments that we're proud of

We got everything functional and running according to the plan! Also, the final design of the website looked a lot better than what we expected. We plan to further build upon the project and use it ourselves for school, and share it with friends and family to help them and receive their feedback as well.

What we learned

In this project, we learned how to use Openai's API to send requests to gpt-turbo-3.5. Additionally, we learned how to use the Blacksheep package to set up an API backend, Hugging Face to download machine learning models, threlte to render 3d components on a website, and pydub to manipulate audio files. We also learned how to collaborate well as a team to create a final product we were all happy with.

What's next for Eloquence

While Eloquence currently supports a variety of features, we’re looking to add even more to allow users to further improve their speaking skills! These features include more context-specific feedback through user input providing the purpose, the time limit, and other helpful information. We also plan to implement a confidence-analysis AI model based on the user's tone to further tailor the feedback. In the future, Eloquence will also grow by allowing users to save their progress, review previous feedback, and keep track of their improvement.

Built With

• blacksheep
• huggingface
• openai
• pydub
• python
• skeletonui
• svelte
• svelte-kit
• threlte
• typescript