How shobhitkori/voice-to-code works
Project type and purpose
• This is a full-stack web application called “Voice-to-Code” that enables users to upload or record audio, transcribe their speech into text, and convert that text into executable source code.
• The root README, plus two parallel “client” directories and a “backend” directory, make it clear that there is a front-end web interface built in Next.js/React/TypeScript and a back-end service in Python using machine-learning models (Whisper for transcription, CodeT5 for code generation).
File structure
• Top-level folders:
• backend: Python service with main.py (FastAPI/Uvicorn entry point), codet5_model.py, whisper_model.py, utils.py, ffmpeg_test.py, Dockerfile, requirements.txt
• client and client-2: two nearly identical Next.js TypeScript apps (likely a v1/v2 layout) with app/ pages, components/ (AudioUpload.tsx, theme provider, UI library), tailwind and PostCSS configs
• Configuration/build files:
• backend/Dockerfile, .gitignore, requirements.txt
• client-*/package.json, tsconfig.json, next.config.ts, tailwind.config.ts, eslint.config.mjs
• No persistent database or external data store is present—state is passed in memory/request.
README for additional insights
• The Overview section details the core workflow: user speaks or uploads audio → backend processes audio → speech-to-text (Whisper) → text-to-code (CodeT5) → result returned to UI.
• Dependencies: Python, FastAPI (implied by uvicorn), React/Next.js/TypeScript, Tailwind CSS.