FunClip: an automated video clipping tool that extracts segments based on ASR transcripts, speaker IDs, or LLM analysis

What it solves

FunClip simplifies the process of extracting specific segments from long videos. Instead of manually scrubbing through a timeline, users can find and clip video sections based on transcribed text, specific speakers, or AI-driven content analysis.

How it works

The tool first performs Automatic Speech Recognition (ASR) on a video using models like Paraformer, Fun-ASR-Nano, or SenseVoice to generate a transcript with timestamps. Users can then clip the video in three ways:

1. Text-based: Selecting specific text segments from the transcript.
2. Speaker-based: Using the CAM++ model to identify speakers and clip segments belonging to a specific person.
3. AI-driven: Integrating LLMs (like Qwen or GPT) or video-understanding models (like TwelveLabs Pegasus) to analyze the transcript or visual content and automatically suggest timestamps for clipping.

Who it’s for

It is designed for content creators, editors, and researchers who need to quickly isolate highlights or specific quotes from video recordings without manual editing.

Highlights

• Industrial-grade ASR: Integrates Paraformer-Large for high-accuracy Chinese speech recognition.
• Smart Clipping: Supports LLM-based inference to automatically identify relevant video segments via prompts.
• Speaker Diarization: Ability to recognize and clip segments based on speaker IDs.
• Multilingual Support: Capable of processing English, Japanese, and various Chinese dialects.
• Hotword Customization: Allows users to specify entity words or names to improve ASR accuracy.
• Visual Understanding: Optional integration with TwelveLabs Pegasus for clipping based on visual events rather than just audio.