Paste a YouTube, TikTok, or Reddit link — or drop an audio file — and watch the clip get cleaned, isolated, and diagnosed in real time. Every stage of the pipeline is shown as it happens.
Runs entirely on your machine. The clip is cleaned with the
same cascade used to build the training corpus, so it's processed exactly like a
training clip — then scored by calibrated linear heads on frozen CLAP embeddings.
See docs/MODEL_CARD.md for the measured, by-video cross-validated accuracy.