VideoLingo

Why we included this project

VideoLingo takes a raw video file and turns it into a fully localized version with clean, single-line subtitles and optional dubbed audio, which makes it a practical tool for anyone who regularly republishes or shares video across language barriers. The whole chain runs in one pass: it downloads the source, transcribes with word-level alignment, splits the text into readable subtitle lines, translates with a multi-step reflect-and-adapt approach, and can generate a dubbed track through providers like GPT-SoVITS, Azure, or OpenAI. Small teams and individual creators get Netflix-style output without hiring a subtitle editor or voice actor. The Streamlit interface starts with one click, supports pausing and resuming long jobs, and works with OpenAI-compatible LLM APIs or fully local setups via Ollama and Edge-TTS. The main trade-off is that transcription quality depends on the chosen WhisperX model, and dubbing is limited to a single speaker per video.