ClarityAI

ClarityAI is like combining Grammarly with the Apple AirPods Pro 3 Live Translation feature — but instead of translating languages or correcting writing, it “translates” accented or unclear English speech into clear, natural-sounding English in real time during Zoom calls.

Non-native speakers talk normally in their own accent; ClarityAI processes their voice live, improves pronunciation, clarity, pacing, and filler words, and sends a clean, native-level English version to the listener while keeping their original voice identity.

Just like Grammarly fixes your writing instantly, ClarityAI fixes your speaking instantly — in real time. This gives non-native speakers a massive advantage in meetings, interviews, sales calls, and presentations without needing coaching or practice.

Technically, it orchestrates multiple modalities — voice input, browser UI, cloud LLM reasoning, and custom rule-based analysis — into one cohesive agent loop. This combination is both innovative and impactful: it brings Grammarly-style intelligence to live conversations, helping people communicate more clearly and confidently. The real-world impact is broad, from better sales calls and interviews to stronger team communication.

Theme Alignment & Technologies Used:

ClarityAI aligns with the theme by merging browsers, voices, clouds, and tools into a single agent:

• Browsers: Floating UI overlay for live transcript and suggestions
• Voices: Real-time mic capture with WebRTC
• Clouds: OpenAI Realtime API for transcription, rewriting, and summaries
• Tools: Rule-based detectors, pacing counters, overlay controls, and WebSockets

Technologies:

• 🎤 Real-time Speech Recognition - Uses Mozilla’s native Web Speech API
• ✨ Text Refinement - OpenAI GPT-4o-mini for accent normalization
• 🎭 Voice Cloning - ElevenLabs for maintaining speaker identity
• 🎨 Clean UI - ShadCN-inspired black and white design
• ⚡ Fast Setup - Uses uv for Python package management
• ElevenLabs API
• GPT-4.1/GPT-5 for rewriting & summaries
• WebRTC for audio capture
• JavaScript/TypeScript
• HTML/CSS browser overlay
• Node.js + WebSockets
• Basic regex logic for filler-word & pace detection