SPEECHMAX

Speech practice is dull and people drop off. Many people are anxious about public speaking, professional coaching is expensive, and the existing tools are clinical and feel like homework rather than something to come back to.

A Duolingo-style progression system with drills modelled on deliberate practice. Webcam analysis runs in real time, using MediaPipe for eye contact and posture, the Web Speech API for transcription, and the Web Audio API for pitch detection, so no audio or video leaves the browser. Five mini-games each isolate one weakness.

SpeechMAX is a browser-based speech practice app. It was built in March 2026 by a team of four at UNIHACK 2026, where it placed 5th of 183. It is one of three hackathon results Bruno Jaamaa lists: 1st at Melbourne Hack, 4th at FoundersHack and 5th of 183 at UNIHACK 2026.

The premise was that many people are anxious about speaking in public, that professional coaching is expensive, and that the existing practice tools feel clinical. The team set out to build a speech coach that anyone could open for free.

The app runs in the browser. The user opens it, grants camera and microphone access, and starts talking. It gives feedback on three signals: voice clarity (filler words, words per minute, pitch variation), eye contact (468 facial landmarks plus iris tracking through MediaPipe), and body language (posture alignment and fidget detection from 33 body keypoints).

A 30-second scan scores the speaker on five axes, and SpeechMAX then recommends mini-games for the weakest ones. Filler Ninja catches "ums" and "likes". Eye Lock trains sustained eye contact with the camera. Pace Racer keeps the speaking rate in a target zone. Pitch Surfer rewards vocal variety. Stage Presence coaches open body language and gestures.

Mike, an AI speech coach running on Gemini 2.5 Flash, reads the speaker's scores, game history and progress and gives short pieces of advice. Scores and history sync to Supabase, with Google sign-in or an anonymous guest mode.

Jaamaa owned the backend and infrastructure, which here means the real-time analysis pipeline and the game UX. His three teammates covered the front end, the UI/UX and character design, and planning. The project was not developed further after the event.

The front end is React 19 with TypeScript (strict mode), Vite, and Tailwind CSS 4. State management uses Zustand 5 with localStorage persistence and Supabase sync. Framer Motion handles the animations.

The analysis pipeline runs on the client. MediaPipe FaceLandmarker (WASM, GPU-accelerated) tracks 468 facial landmarks for eye contact quality using a 3-signal fusion approach: iris position, head pose estimation, and gaze direction. MediaPipe PoseLandmarker tracks 33 body keypoints for posture alignment and fidget detection.

Audio processing runs through the Web Speech API for real-time transcription with noise suppression, through a DynamicsCompressorNode (threshold -40dB, 4:1 ratio), into an AnalyserNode for FFT pitch detection via autocorrelation. Filler word detection uses count-based pattern matching against a configurable dictionary.

The backend is Supabase: anonymous and Google OAuth authentication, PostgreSQL with RLS for the profiles, scan_results and game_results tables, and a gemini-proxy Edge Function that routes Gemini API calls through the server so the API key never touches the client.