Transcription Pipeline
Agent skills for transcription in stages: retrieve, preprocess (EQ, denoise, VAD), transcribe (OpenRouter/Gemini, AssemblyAI, local Whisper/Parakeet), clean, label speakers, reshape, Typst PDF, save
About This Project
Transcription-Pipeline is a modular toolset for converting raw audio recordings into structured, usable documentation. It automates the entire lifecycle of a voice note, including retrieval from various sources, audio preprocessing with equalization and silence trimming, multi-provider transcription, speaker labeling, and document formatting via Typst.
Built as a Python-based plugin, the system supports integration with LLM providers like OpenRouter, Gemini, and AssemblyAI, as well as local ASR solutions. It features a library of over 200 system prompts for content transformation, allowing users to reshape raw transcripts into technical specifications, meeting minutes, blog posts, or structured data formats. By maintaining persistent artifacts and clear metadata for every processing step, it provides a reliable, reproducible workflow for developers and researchers who rely on voice-to-text documentation.