What is worldcupvoice?

WorldCupVoice is an AI system that adds live spoken commentary to a video stream, in real time, as the action happens. It watches the same live video feed that viewers see, using a real time communication service called Agora, and generates spoken commentary that plays back into the stream alongside the original broadcast. The World Cup is the showcase example the project is built around, but the same setup works for any live stream. One notable use case the project highlights is accessibility: AI commentary can describe field level detail, such as where the ball is or who is making a run, that human commentators sometimes skip, which can help blind and low vision viewers follow the action. The README is clear that this is meant to add to human commentary, not replace it. Under the hood, a live source such as OBS pushes video into Agora's system over RTMP, which distributes it to viewers and to an AI backend at the same time. The backend samples video frames, uses a vision model to understand what is happening, and turns that into spoken commentary using a text to speech service. Several TTS providers are supported, including OpenAI, ElevenLabs, and Fish Audio, and the project ships sample commentator voice profiles for English, French, and Chinese. Setting this up requires both a frontend web app, built with a JavaScript package manager, and a Python backend server, along with accounts and API keys for Agora and whichever text to speech provider is chosen. There are explicit controls, such as start and stop commands and a session time limit, meant to keep AI usage and cost under control. The project is open source under the MIT license, so it can be freely used and modified, but running it live requires setting up these external accounts yourself.