Inspiration

What it does

How we built it

Speech Mate 🎙️

Speech Mate is an AI-powered public speaking coach designed to help you write, practice, and perfect your speeches. From generating an outline to analyzing your final performance, Speech Mate is your all-in-one partner for confident communication.

Key Features

• AI-Powered Speech Generation: Generate a detailed, structured speech outline in seconds. Just provide a topic, tone, audience, and desired duration, and let our AI build your speech.
• Multimodal Performance Analysis: Upload a video of your speech—along with your slides, notes, or any other materials—and receive a comprehensive, AI-driven report.
• Detailed Feedback: Get an overall score and detailed breakdowns on:
  • Content & Structure: How well you stayed on topic and organized your ideas.
  • Delivery & Body Language: Analysis of your posture, gestures, and eye contact.
  • Vocal Variety: Feedback on your pace, pitch, tone, and intonation.
  • Filler Word Counting: A precise count of "ums," "ahs," "likes," etc.
  • Accent & Clarity: Constructive analysis of your pronunciation and language.
• Actionable Improvement Plan: Receive specific, actionable tips and a list of recommended YouTube videos tailored to your unique weaknesses.
• Text-to-Speech Practice: Listen to your generated speech with a realistic AI voice (powered by ElevenLabs) to practice your timing and cadence.
• Secure Authentication: User accounts are secured via Google OAuth 2.0.

How It Works

Speech Mate is a full-stack application composed of a React frontend and a Spring Boot backend.

1. Authentication: The user logs in using their Google account. The Spring Boot backend uses Spring Security and OAuth2 to authenticate the user and establish a session.
2. Speech Generation:
   • A user fills out the "Generate Speech" form in the React UI.
   • This calls the /api/gemini/generate-outline endpoint on the Spring Boot server.
   • The GeminiService constructs a detailed prompt for the Gemini 2.0 Flash model, asking it to return a structured JSON object containing a title, thesis, timed sections, talking points, and a conclusion.
3. Speech Analysis:
   • A user uploads a video file and/or other documents (PDFs, images).
   • This calls the /api/gemini/analyze-speech endpoint with multipart/form-data.
   • The GeminiService uses the powerful Gemini 2.5 Pro multimodal model. It combines the video, slides, and user-provided context (topic, audience) into a single, comprehensive prompt.
   • This prompt instructs the AI to act as an expert speech coach and return a highly detailed JSON object with scores, content summaries, language/accent analysis, intonation patterns, filler word counts, specific statement-by-statement feedback, and YouTube recommendations.
   • The React frontend (SpeechAnalysis.tsx) parses this complex JSON response and displays it in a user-friendly report.
4. Speech Practice:
   • The user can send text to the /api/elevenlabs/text-to-speech endpoint.
   • The backend calls the ElevenLabs API to generate high-quality audio, which the user can play back to practice their delivery.

Technology Stack

• Backend: Java 21, Spring Boot 3.5.7, Spring Security (OAuth2), Spring WebFlux (WebClient)
• Frontend: React 19, TypeScript, Vite, React Router, Axios, Framer Motion
• AI APIs: Google Gemini (2.5 Pro, 2.0 Flash, 2.0 Flash-Lite), ElevenLabs
• Authentication: Google OAuth 2.0

Built With

• elevenlabs
• gemini
• google-oauth
• java
• react
• springboot
• typescript
• vite