FunClip

Why we included this project

FunClip treats speech recognition as the front end of an editing workflow: it runs FunASR's Paraformer models over a video's audio, transcribes what was said, and lets you rebuild the cut by picking the exact sentences or speakers you want to keep. For people who edit interviews, lectures, or long recordings, that means clean SRT subtitles and a fast way to drop filler or mistakes without scrubbing the timeline. Speaker recognition helps isolate one person's voice from multi-person footage, and hotword customization sharpens accuracy on proper names and domain terms. Everything runs locally in a Gradio interface, which matters when source footage can't leave your machine, and the optional LLM mode lets you describe the segment you want rather than clicking through lines.