TalkStar

• 

  Speech Upload History

• 

  Suggestions

• 

  Speech Graph

• 

  Speech Transcript

Inspiration

Public speaking can be daunting, but we believe everyone has the potential to deliver impactful speeches with the right tools. Our goal was to create a platform that helps individuals—students, professionals, or everyday speakers—enhance their delivery with real-time, data-driven feedback. By combining quantitative speech analysis with AI-powered insights, we aimed to provide quick, actionable suggestions that empower users to speak with confidence and clarity.

What it does

TalkStar lets you upload your speech or presentation and receive detailed, actionable feedback. It analyzes key data points such as filler word usage, qualities in the voice recording, then presents these insights in a digestible format with graphs along with personalized suggestions. Plus, it tracks your progress over time so you can see how you improve.

How we built it

We built TalkStar using a mix of cutting-edge technologies:

• Backend:
  We ran the librosa library on a python server for quantitative speech analysis to extract metrics like pitch, jitter, and shimmer. We also integrated the AssemblyAI API and Claude-AI to provide advanced feedback and format the suggestions.

• Frontend (React):
  Our interactive user interface is built with React, allowing users to easily upload speeches, view real-time analysis, and track their progress.

• File Storage:
  We implemented Pinata for fast and reliable storage of uploaded speeches, including various other metadata such as date and duration.

Challenges we ran into

The ideation phase was especially challenging as we worked to create a service that truly helps people become rockstars in their own right. Integrating multiple APIs and ensuring smooth data flow between our Python backend and React frontend demanded iterative problem-solving. Additionally, translating raw quantitative data into actionable, understandable feedback required careful calibration.

Accomplishments that we're proud of

We’re proud of developing a powerful tool that delivers real-time, actionable feedback on speech performance. Successfully merging independent quantitative analysis with advanced AI interpretation—and enabling users to track their progress over time—is a significant milestone that sets TalkStar apart.

What we learned

We learned how to creatively integrate multiple APIs to generate feedback that surpasses standalone AI. By building a Python server backend as an API, we ensured seamless interaction with our React frontend. Exploring AssemblyAI and Claude-AI deepened our understanding of extracting meaningful speech data, while Pinata file storage enabled fast and reliable storage of user-uploaded audio files. Most importantly, we learned to keep our scope limited in order to ensure we had a polished product at the end.

What's next for TalkStar

Looking ahead, we plan to expand TalkStar’s analytics capabilities, introduce personalized coaching features, and enhance community-driven feedback. Our goal is to refine our metrics further and improve the overall user experience so that every user can truly become a confident, impactful speaker.

Built With

• assemblyai
• claude
• librosa
• python
• react