OpenBiliClaw Media Processing & Video Knowledge Miner

OpenBiliClaw Media Processing & Video Knowledge Miner

@whiteguo233/openbiliclaw·v1.1.2·tools

A video intelligence and media knowledge extraction pipeline for DeepSeek Harness. Extracts high-fidelity audio transcripts, maps danmaku sentiment spikes, and captures keyframe slides.

Video-Transcribe · Whisper · Keyframe-Extraction · Media-Knowledge

GitHubNPM
Installation & Integration
bash
manager:
$ pnpm add @whiteguo233/openbiliclaw

Architecture

OpenBiliClaw unlocks video and media archives for autonomous knowledge consumption. The vast majority of cutting-edge tech conferences, developer tutorials, and academic webinars are published exclusively as long-form videos inaccessible to text agents. This plugin integrates media pipelines directly into DeepSeek Harness. Agents extract audio streams for timestamped Whisper transcription, sample visual keyframes of slide transitions, and analyze crowd sentiment spikes via live danmaku comment density. Long videos distill into structured, citation-backed Markdown study guides in minutes.

Core Features

01

Rapid Multi-Hour Media Transcription: Precise transcription for developer webinars and talks

02

Intelligent Slide Keyframe Extraction: Automatically captures crisp presentation slides

03

Danmaku Crowd Sentiment Analysis: Identifies viral highlights and technical breakthrough moments

04

Clickable Timestamped Markdown: Notes link directly back to video playback timestamps

Core Workflow

01

Media Demuxing & Audio Rip

Demuxes video containers into clean audio streams split into processing chunks.

02

Timestamped ASR Transcription

Runs Whisper ASR to generate timestamp-accurate text transcripts.

03

Slide Keyframe Detection

Identifies scene transitions to capture sharp presentation slide keyframes.

04

Markdown Knowledge Digest

Synthesizes prose, slide visuals, and time-linked anchors into clean Markdown wikis.

Configuration Parameters Reference (YAML / JSON)

ParameterTypeDefaultDescription
asrModelstringwhisper-baseTarget Whisper speech-to-text model tier
extractSlidesbooleantrueAutomatically detect and extract presentation slides
minDanmakuPeakRationumber2.5Danmaku spike threshold multiplier for highlights