@Ryrenz: Awesome, drop the recording and it directly generates notes, fully offline. It has 2.2K stars on GitHub

A win-win: throw in a recording and get notes directly, completely offline. With 2.2K stars on GitHub, if you listen to a two-hour podcast or online course, there are usually only a few key segments you really want to keep. Manually taking notes requires pausing and rewinding, while online transcription tools require uploading audio, which feels risky when dealing with company meetings or private content. AudioNotes runs entirely locally. Upload an audio or video file, and it first generates a transcript, then organizes it into Markdown notes. After that, you can ask questions directly based on the content, with the conversation history saved as well. It supports browser recording and domain-specific hotwords, allowing you to correct professional term transcription errors. The reason it can run fully locally is that speech recognition uses FunASR and the language model runs on Ollama, with the default configuration being FunAudioLLM/Fun-ASR-Nano-2512 and qwen3.5:2b. Audio/video files, transcripts, notes, and conversation history are all saved locally by default, and the web service only listens on the local address. Deployment options include Docker and Python 3.12 source code, with a default port of 15433. The Docker version uses admin for both the initial username and password; remember to change them after installation. The first use requires an internet connection to download images and models, with the speech model being approximately 2.15GB. Only locally runnable tools can be trusted to handle truly important content.

GitHub: — # harry0703/AudioNotes Source: https://github.com/harry0703/AudioNotes # AudioNotes AudioNotes is a locally running audio/video transcription and intelligent note-taking tool, ideal for organizing meetings, interviews, courses, and voice memos. Key features: - Upload audio or video to automatically generate transcripts. - Organize recognition results into clear Markdown notes. - Continue asking questions based on audio/video content. - Supports browser recording, domain-specific hotwords, and conversation history. ## Local-First and Privacy AudioNotes defaults to performing speech recognition, note organization, and content Q&A on your computer: - Audio/video files, transcripts, notes, and conversation history are stored locally. - FunASR recognizes speech locally, and Ollama runs the large language model locally. - By default, audio/video content is not sent to third-party AI APIs. - The web page is only accessible locally by default and not open to the LAN or internet. The first installation still requires an internet connection to download the application image and models, but processing personal content does not rely on cloud AI services. Please use it on a trusted computer and properly protect your local account and disk data. ## Demo ### Audio/Video Recognition and Organization Audio/Video Recognition and Organization ### Dialogue with Audio/Video Content Dialogue with Audio/Video Content ## Docker Quick Start Before starting, please install and launch: - Docker Desktop (https://www.docker.com/products/docker-desktop/) - Ollama (https://ollama.com/download) It is recommended to allocate at least 12GB of memory to Docker Desktop; for longer audio/video processing, 16GB is recommended, with approximately 10GB of free disk space. ### 1. Prepare Local Model bash ollama pull qwen3.5:2b ### 2. Download and Start AudioNotes bash mkdir audionotes cd audionotes curl -fsSL https://github.com/harry0703/AudioNotes/raw/main/docker-compose.yml -o docker-compose.yml curl -fsSL https://github.com/harry0703/AudioNotes/raw/main/.env.example -o .env docker compose up -d --remove-orphans ### 3. Open the Page Visit . The default username and password are both admin. The first recognition will automatically download the approximately 2.15GB speech model; the wait time depends on network speed. The model will be saved locally after download, so no need to re-download it subsequently. ## Start from Source (Optional) If you prefer to run the source code directly, please prepare Python 3.12, ffmpeg, and Ollama first. AudioNotes will automatically create local data files, so no database service installation or configuration is needed. bash git clone https://github.com/harry0703/AudioNotes.git cd AudioNotes python3.12 -m venv .venv source .venv/bin/activate python -m pip install -r requirements.txt cp .env.example .env ollama pull qwen3.5:2b Once Ollama is confirmed to be running, execute: bash python -m chainlit run main.py --host 127.0.0.1 --port 15433 Also visit . To exit the virtual environment, run deactivate. ## Usage 1. After logging in, upload audio or video, or use browser recording directly. 2. Wait for speech recognition and note organization to complete. 3. View the full transcript and structured notes. 4. In the input box, continue to ask about responsible persons, time, conclusions, or other details. 5. Reopen previous notes from the history on the left. ## Default Models - Speech Recognition: FunAudioLLM/Fun-ASR-Nano-2512 - Note Organization and Q&A: qwen3.5:2b The default combination prioritizes Chinese recognition quality and resource usage on ordinary computers. Recognition results may still be affected by recording quality, environmental noise, accents, and overlapping speech; please verify important content against the original audio. To recognize more languages, open .env, change ASR_MODEL to FunAudioLLM/Fun-ASR-MLT-Nano-2512, and then restart the service. The default model does not change automatically, and the current page will not switch models during runtime. ## Data Management When using Docker, uploaded files, FunASR model cache, logs, and conversation history are saved in the created audionotes folder. When using the source code, personal data is saved in the project’s storage folder, and the FunASR model is stored in the local model cache. Ollama models are managed separately by Ollama. - Backup: Stop the service and copy the entire audionotes folder. - Migration: Copy the entire folder to a new computer, install Docker and Ollama, pull the default model, and then start the service. - Cleanup: After confirming that historical data is no longer needed, stop the service and delete the folder. When using Docker, to stop or restart: bash docker compose down docker compose up -d When using Docker, to update to the latest version: bash docker compose pull docker compose up -d --remove-orphans ## FAQ If the page cannot open when using Docker, first confirm that both Docker Desktop and Ollama are running, then check the service status: bash docker compose ps docker compose logs -f webui If the source code fails to start, please check the error messages in the terminal first and confirm that Python version, ffmpeg, and Ollama are all working properly. If prompted that the model is not found, re-run: bash ollama pull qwen3.5:2b If the page interrupts or the service restarts when processing long audio, please increase the memory quota for Docker Desktop. In practice, 8GB may be insufficient; 16GB is recommended. If the computer is shared by multiple people, open the .env file and modify USERNAME and PASSWORD, then restart the service. Linux computers with NVIDIA graphics cards and already installed NVIDIA drivers and Container Toolkit can use the optional configuration to enable GPU recognition acceleration: bash git clone https://github.com/harry0703/AudioNotes.git cd AudioNotes cp .env.example .env docker compose -f docker-compose.yml -f docker-compose.gpu.yml up -d --build The default quick start still uses the CPU image and does not require an NVIDIA graphics card. Building the GPU image for the first time requires downloading large CUDA dependencies.