Matladi Chudu

Speak•Analyze•Improve•Repeat

Comment

Inspiration

Matladi Chudu ("Speak and See") was born from the need for a private, judgment-free space to practice speaking—whether for public speaking, language learning, interviews, or personal clarity. Many existing tools require internet, send data to the cloud, or feel clinical. We wanted a beautiful, fully offline-first PWA that lets anyone record, review, and reflect on their voice with meaningful insights, while keeping everything on their device.

What it does

Matladi Chudu is a privacy-first speaking practice app. Users record 5-minute calm practice sessions, get automatic transcription (Deepgram when online, browser fallback offline), and receive instant offline analysis including:

Clarity score, pause ratio, and energy level Filler word detection Pronunciation scoring and weak word highlights Vocabulary diversity and advanced word tracking Session history, streak counter, and progress graph Manual transcript editing, reflections, and exportable reports (JSON + clean text)

All audio and data stay on the user’s device. It works as a standalone PWA.

How we built it

Frontend: Pure HTML, CSS (custom design with Fraunces + Manrope), and vanilla JavaScript. Audio: MediaRecorder API + Web Audio API for real-time visualizer and DEEPGRAM API for Analysis and Confidence score Transcription: Deepgram Nova-2 API (with robust offline fallback using Web Speech API). Analysis: Custom algorithms for silence detection, RMS energy, filler words, vocabulary, and pronunciation confidence. Storage: localStorage for sessions and reflections. PWA: Service Worker for offline caching + full manifest support. Export: JSON metadata and formatted text reports.

Challenges we ran into

Balancing accurate real-time analysis with full offline capability. Handling inconsistent browser support for SpeechRecognition across devices. Creating meaningful metrics from raw audio without heavy ML models running locally. Keeping the UI calm, elegant, and motivating rather than overwhelming with data. Ensuring Deepgram integration is graceful when offline or rate-limited.

Accomplishments that we're proud of

A genuinely beautiful, calm, and functional speaking coach that works completely offline. Smart hybrid transcription (Deepgram + fallback) with pronunciation and vocabulary insights. Clean, professional PWA experience with smooth audio visualizer and progress tracking. Full local data privacy — no accounts, no tracking, no cloud dependency for core use. Exportable professional reports for self-review or sharing.

What we learned

We learned how powerful modern browser APIs have become for media and speech processing. We also discovered that thoughtful, minimalist design combined with practical feedback can make practice feel less like "work" and more like mindful growth. Privacy and offline capability are not just features — they fundamentally change how people feel about using a tool.

What's next for Matladi Chudu

Improved local speech models (WebAssembly-based) for better offline accuracy. Script/prompt library for guided practice. More advanced pacing, intonation, and filler pattern analysis. Dark mode and mobile-first refinements. Optional anonymous aggregated insights (opt-in) to help improve the app. Template-based reflection prompts and goal setting.

Built With

Share this project:

Updates