Turn audio ramble into structured Markdown, offline

What it does

AudioNotes ingests audio or video files, transcribes the speech with FunASR, and passes the text to a local Qwen2 model via Ollama. The LLM formats the raw transcript into structured Markdown—headings, bullets, and all—so you get a readable summary instead of a wall of text. It also exposes a chat interface that lets you ask questions against the processed content.

The interesting bit

It is largely glue code—FunASR plus Ollama wrapped in a Chainlit UI—but the value is in the wiring: it keeps the entire pipeline on-premise and hands you a formatted document rather than a raw subtitle file.

Key highlights

• Fully local pipeline: FunASR for speech recognition, Qwen2 (via Ollama) for summarization and structuring.
• Output is structured Markdown, not just a transcript dump.
• Includes a conversational interface to query the extracted content after processing.
• Packaged as a Docker Compose app with a web UI; local deployment requires PostgreSQL.
• Built in Python using Chainlit for the frontend.

Caveats

• The README is thin on architecture details—how chunking, context limits, or long-file handling work is unclear.
• Ships with hard-coded admin/admin credentials, so treat it as a local-trust tool rather than a multi-user service.

Verdict

Worth a look if you need a private, self-hosted way to turn recordings into meeting notes or lecture summaries. Skip it if you are after a managed SaaS with mobile sync and enterprise SSO; this is a bring-your-own-hardware Python stack.