harry0703/AudioNotes

What it solves

AudioNotes provides a private, local solution for converting audio and video recordings into structured text and intelligent notes. It eliminates the need to send sensitive meeting, interview, or lecture data to cloud-based AI services, ensuring privacy while automating the transcription and summarization process.

How it works

The tool runs locally using a combination of specialized models. It uses FunASR for speech-to-text recognition and Ollama to run Large Language Models (like qwen3.5:2b) for organizing the transcripts into Markdown notes and answering follow-up questions about the content. It can be deployed via Docker or run directly from source code.

Who it’s for

It is designed for individuals who need to organize meetings, interviews, courses, or voice memos but require high data privacy and prefer to run their AI tools on their own hardware.

Highlights

• Local-First Privacy: Audio, transcripts, and notes are stored on the local machine; no data is sent to third-party AI APIs by default.
• End-to-End Workflow: Handles everything from browser-based recording and file upload to transcription and structured note generation.
• Interactive Q&A: Allows users to ask specific questions about the content of the uploaded audio or video.
• Flexible Deployment: Supports both CPU-based Docker setups and GPU-accelerated configurations for Linux users with NVIDIA cards.