Voice-to-Note

Turn spoken audio into study material. Voice-to-Note transcribes an audio clip with OpenAI Whisper, then uses a GPT chat model to convert the transcript into flashcards (Front::Back pairs) you can review in the UI.

Hackathon project. The backend is functional; the frontend is mid-refactor and does not currently build (see Status below).

Preview

TODO — add a screenshot once the frontend build is fixed. The intended UI is a Home page (record/upload audio) and a /flashcards review page.

Features

• Audio → transcript — POST /api/transcribe accepts base64 audio, sends it to the Whisper API (whisper-1), and returns the transcript.
• Transcript → flashcards — the transcript is passed to a GPT chat model (gpt-3.5-turbo) that returns short Front::Back flashcard strings, parsed into structured cards.
• No-key demo mode — if no OpenAI key is configured, the transcribe endpoint returns a canned demo transcript and sample flashcards, so the flow can be demonstrated without credentials.
• Local persistence — generated notes and sentences are saved to local JSON files (notes.json, local_sentences.json) and served back via /api/notes and /api/flashcards.

Tech stack

Layer	Choices
Frontend	React 19, Vite 7, React Router 7
Backend	Node.js, Express 5, node-fetch, form-data
AI	OpenAI Whisper (transcription) + GPT chat (flashcard generation)

API (backend)

Method	Route	Description
POST	/api/transcribe	{ audio_base64, filename?, mimetype?, save? } → { transcript, generated }
POST	/api/process-text	{ text } → flashcards parsed from raw text
GET	/api/notes	Returns saved notes
GET	/api/flashcards	Returns generated flashcards

Getting started

Prerequisites: Node.js 18+ (the backend uses global fetch/FormData).

Backend (works today)

cd backend
npm install
node server.js        # http://localhost:3001

Without an API key it runs in demo mode. To enable real transcription, set an OpenAI key (see Environment variables).

Frontend (currently broken — see Status)

cd frontend
npm install
npm run dev           # Vite dev server (once the build issues are resolved)

Environment variables

Backend .env (gitignored — do not commit). Variable names only; the server accepts any one of:

Variable	Used for
OPENAI_API_KEY (preferred) / OPENAI_KEY / OPENAI	OpenAI Whisper + chat completion access

Status / known issues

This project is in progress. Two issues block the frontend from building:

1. Merge conflict in frontend/package.json — the file still contains <<<<<<< HEAD … ======= … >>>>>>> frontend markers from an unresolved merge and is not valid JSON. It needs to be resolved (keep the voicetonotes app config with React/Vite deps).
2. Missing component — frontend/src/App.jsx imports ./pages/Home, but no pages/Home file exists in src/ (only header.jsx, flashcard.jsx, flashcardPage.jsx). The Home page needs to be added or the import repaired.

The backend has no known blockers and runs standalone.

Project structure

voice-to-note/
├── backend/
│   ├── server.js        # Express API (transcribe, notes, flashcards) + demo fallback
│   ├── processing.js    # Transcript → flashcard parsing + local JSON persistence
│   └── harvard.wav      # Sample audio for testing
└── frontend/
    └── src/
        ├── App.jsx           # Router: / and /flashcards
        ├── flashcardPage.jsx # Flashcard review page
        ├── flashcard.jsx     # Flashcard component
        └── header.jsx

Technical highlights

• Graceful degradation. The transcribe endpoint detects a missing API key and falls back to a demo response — a deliberate choice so the project always demos, even on a judge's machine with no credentials.
• Tolerant flashcard parsing. parseLineToFlashcard accepts several formats (Front::Back, Front: Back, "X is Y") so messy model output still maps to a usable card.

License

No license file is present; treat as all rights reserved unless one is added.