What problem does it solve? Legal professionals handle sensitive recordings—client meetings, hearings, depositions, oral arguments, WhatsApp audio—that must be converted to text without leaking confidential content to cloud services. This Skill transcribes recordings entirely offline on the local machine, producing timestamped, citable text while preserving professional secrecy. ## Core Features & Use Cases - Offline transcription: Converts any audio or video file into a .srt subtitle file and a clean .md transcript with [HH:MM:SS] timestamps, using whisper.cpp locally so no audio leaves the machine. - Resumable processing: Transcribes in 10-minute blocks with incremental saving, so an interrupted run resumes where it stopped when the command is re-executed. - Speaker labeling and evidence extraction: Optionally replaces [?] speaker placeholders with real roles (judge, attorney, witness, client) based on context references, and extracts legally relevant statements into a timestamped evidence summary. - Use Case: After a two-hour hearing, run the Skill on the recording to get a full transcript, then label each speaker using the hearing's ata and build a PROVA_ORAL.md file citing the exact timestamps of admissions and contradictions. ## Quick Start Transcribe the hearing recording at casos/Silva-vs-Maria/autos/audiencia.webm and identify the speakers using the case's ata.
