How cyberytti/video2transcript works
Project type and purpose
• This is a full-stack, monolithic Flask web application whose goal is to let users upload MP4 videos, convert them into audio, transcribe speech using an AI model (Whisper via the Groq API), translate the transcript into multiple languages (using Gemma3 via Ollama), and download the result as a JSON file.
File structure
• Top-level:
• app.py – Flask application and route definitions
• main.py – Orchestrator for the processing pipeline
• functions/ – Pure-Python modules implementing each processing step
• templates/ + static/ – Frontend HTML, JS, CSS
• outputs/ – Stores generated transcript JSONs
• requirements.txt, README.md, LICENSE
• The “functions” directory reveals a simple pipeline: video_to_wav → wav_to_16kwav → convertedwav_to_transcript → transcript_lan_covert.
Extract additional insights from the README
• Core external dependencies: FFmpeg (local CLI), Groq API (remote HTTP) for speech recognition, Ollama with Gemma3 (local HTTP) for translation.
• UI is a single-page form with progress indicators and a “Download JSON” link.
Key details to capture
• All four processing steps in order
• The role of FFmpeg as an external binary
• Two distinct external AI services (Groq vs. Ollama)
• The persistence of uploads and outputs on disk
• The synchronous request-response pattern in Flask