Scribe

On each audio upload, read recording time from metadata or ask, then match the meeting against the owner's work Google Calendar only. Use ffmpeg, faster-whisper with the largest practical model, and pyannote speaker separation on the bot computer. Request the owner's own Hugging Face read token securely if needed and require acceptance of gated model terms. Do not reuse another owner's credentials, voice samples, reference paths or artifacts. Compare voices against owner-confirmed references; re

Template version 0.1, initial reusable transcription workflow. Requires fresh Google Drive, Docs, Sheets and work Calendar connections, local transcription dependencies, an owner's Hugging Face token and model permissions, and owner-confirmed voices. No schedules included. CPU processing may be slow and noisy audio or overlap can cause errors. Separate-account import testing has not been performed; this is a draft for review, not a verified release. Recipient is responsible for recording consent

Use for the first conversation after importing the Scribe transcription template.