What problem does it solve? Bilibili videos lack readable text versions of their spoken content, and raw ASR output arrives as continuous text without punctuation, full of homophone errors. This Skill turns a Bilibili video URL into a punctuated, proofread transcript, including bilingual dual-audio videos where English original and Chinese voice-over are mixed in one track. ## Core Features & Use Cases - Silence-aware chunked ASR: Transcodes MP4 to 16kHz WAV, detects natural pause boundaries, partitions audio with a DP-optimal chunking strategy, and transcribes chunks in parallel against a Whisper-compatible endpoint. - LLM transcript cleanup: Chunks raw ASR text, fixes homophones and adds punctuation via an OpenAI-compatible endpoint, with per-chunk length validation, an escalation ladder for repetition-loop failures, and resumable state markers. - Bilingual dual-pass reconciliation: Re-runs ASR with an English language hint, aligns both timelines into ~90s windows, and merges them into an interleaved zh/EN transcript with divergence flags. - Batch channel processing: Enumerates an uploader's videos into a manifest and runs a crash-safe producer/consumer queue where downloads overlap ASR work. - Use Case: Point the Skill at a 90-minute Mandarin tech talk on Bilibili; it downloads the audio, transcribes it in parallel chunks, flags low-confidence segments, and produces a cleaned, paragraph-formatted transcript. ## Quick Start Transcribe this Bilibili video into a cleaned transcript: https://www.bilibili.com/video/BV1xx/
