AudioNotes
https://github.com/harry0703/audionotes- Category
- Productivity
- Rank
- No. 964Tools index
Previous survey · No. 948 ·
- Pricing
- Open Source
- Type
- TOOL
- Builder
- harry0703
- GitHub
- 2.5k stars
- Date
About
Quickly extract audio/video content and organize it into structured Markdown notes — transcribe, summarize, and outline in one pass.
What it does
AudioNotes runs a private media workspace on your computer. Upload a recording or capture audio in the browser, then inspect the transcript and organized notes. Its chat keeps the recording context available for questions about decisions, owners, dates, and other details. Previous sessions remain accessible through local history.
Why it's ranked here
This is a strong choice when local processing matters more than a lightweight setup. Speech recognition, note generation, question answering, media, and history stay on the machine by default. The code also separates transcription failures from model failures, preserving completed text when note generation breaks. Substantial memory, storage, and model downloads temper that appeal.
What's good
Long transcripts receive staged summarization before final note generation, reducing context-window pressure. Prompts explicitly preserve names, numbers, dates, conclusions, limits, tasks, and unresolved questions while rejecting unsupported additions. Uploaded media remains playable in reopened sessions. Logs omit transcript bodies and user questions, and microphone recordings are deleted after processing.
Tradeoffs
Setup requires Docker Desktop and Ollama, or Python 3.12, ffmpeg, and Ollama from source. The recommended container allocation starts at 12GB of memory, rises to 16GB for longer media, and needs about 10GB of disk space. Initial speech-model download is about 2.15GB. Recognition can degrade with noise, accents, poor recordings, or overlapping speakers. Default credentials require replacement.
How to use it well
Use it for sensitive meetings, interviews, courses, and voice memos when a trusted computer can hold the complete record. Verify important claims against the original audio, then use follow-up questions to locate decisions and obligations. Add domain hotwords for specialized vocabulary. Do not treat it as a turnkey shared cloud workspace or expose the default local deployment publicly.
Technical notes+
main.py implements the Chainlit web workflow, password authentication, media previews, browser recording, staged error handling, and streamed responses. app/services/asr_funasr.py wraps FunASR, moves synchronous inference to a worker thread, and serializes access with an async lock. app/services/ollama.py uses Ollama through an OpenAI-compatible client and recursively condenses oversized summaries for up to four merge rounds. app/services/data_layer.py provides SQLite-backed Chainlit history plus local attachment storage, path validation, WAL mode, foreign keys, and busy timeouts. docker-compose.yml binds the host port to loopback, drops Linux capabilities, persists models and storage, and connects the container to host Ollama. tests/test_core.py covers workflow structure, deployment configuration, model options, subtitle behavior, and error distinctions.
Observed
- License
- MIT License
- Primary language
- Python
- Install surface
- Docker Compose image or Python 3.12 source installation from pinned requirements
- Interface
- Chainlit web interface with audio and video uploads, browser recording, history, and contextual chat
- Local storage
- SQLite stores application history, while uploaded media and related data use local filesystem storage
- Model services
- FunASR performs speech recognition and Ollama provides note generation and question answering
- Hardware support
- CPU is the default; an optional Linux NVIDIA GPU configuration enables CUDA recognition
- Tests
- The repository includes a unittest-based tests directory
Read from README.md, requirements.txt, main.py, app/utils/utils.py, app/services/ollama.py, app/services/asr_funasr.py, app/services/data_layer.py, tests/test_core.py, LICENSE, Dockerfile, chainlit.md, .env.example, chainlit_zh-CN.md, docker-compose.yml.
What it can do
Transcribe audio content to text
Audio file → Text transcription
Transcribe video content to text
Video file → Text transcription
Summarize audio/video content
Audio or video file → Summary text
Create structured outlines from audio/video
Audio or video file → Organized outline
Convert audio/video to Markdown notes
Audio or video file → Structured Markdown document
Extract key content from audio recordings
Audio file → Extracted key points
Extract key content from video recordings
Video file → Extracted key points
Tags
Tech Stack
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.