Picard-Multilingual-Auto-Tagger

MusicBrainz Picard docker integration (plugin) for generating non-Latin track titles using a local LLM. For example arabic song title "al sahra" ->transliterate "restoring" to "الصحراء" another exapmle japanese song title "Ue o Muite Arukō" transliterate to "上を向いて歩こう"

I originally built this for Arabic music, but the same idea could potentially be adapted for other languages and writing systems.

This is an experimental community project and is not an official MusicBrainz Picard plugin.

Why I made this

I use MusicBrainz Picard to organize my music library, but I often ran into the same problem: MusicBrainz may only have the English or Latin-script title for a track, even when the song is originally known in Arabic or another local language.

Doing this manually for every track takes a lot of time.

This project connects Picard to a small companion web app. The app receives the selected album and track metadata, sends it to a local Ollama model, and suggests localized titles that can be copied back into Picard.

How it works

1. Right-click an album or track inside Picard.
2. Select the multilingual tagging plugin.
3. A small browser window opens from inside the Picard container.
4. The album, artist and track fields are populated automatically.
5. Click Generate to send the information to the local LLM.
6. click send to send the song title to the specific track automatically.
7. When a suggestion is incorrect, edit it manually and click Verify & Save.

Verified corrections are stored in the local SQLite database. The next time the same title appears, the saved result can be reused instead of asking the LLM again.

Screenshots

Open the plugin from Picard

Metadata is automatically populated in the companion app

Generate localized titles

Click send to automatically send the song title back to picard

auto-tagger can also be used alone from web browser

Current features

• Works from the Picard album and track context menu
• Automatically transfers selected Picard metadata to the companion app
• Uses a local Ollama model
• Supports album-level and individual-track generation
• Stores manually verified corrections in SQLite
• Reuses saved corrections without calling the LLM again
• Runs through Docker Compose
• Designed around the jlesage/musicbrainz-picard container
• send button, sends only that title back to its matching track in Picard and it dosent save a user must save from picard (this is done for safety)

Project structure

.
├── auto-tagger/
│   ├── app/
│   │   ├── main.py
│   │   └── picard_prefill.py
│   ├── Dockerfile
│   └── requirements.txt
│
├── picard-plugin/
│   ├── arabic_tagging.zip
│   └── plugin_source/
│       └── arabic_tagging/
│           └── __init__.py
│
├── compose.yaml
└── compose.picard-example.yaml

Some internal names still use arabic-sort or arabic_tagging because the project was originally created specifically for Arabic. These names can be cleaned up in a future version.

Requirements

• Docker and Docker Compose
• MusicBrainz Picard
• The jlesage/musicbrainz-picard Docker image
• Ollama
• A local Ollama model such as qwen3:8b the app have instruction to not use reasoning
• Enough memory to run the selected model

The supplied Compose file runs Ollama as part of the stack.

Installation

Choose the installation method that fits your setup:

• Docker Compose / command line: continue with the instructions below.
• Arcane: see the step-by-step Arcane installation guide.

Check both provided Compose files before starting:

• compose.yaml
• compose.picard-example.yaml

1. Clone the repository

git clone https://github.com/ForeignWelcome/Picard-Multilingual-Auto-Tagger.git
cd Picard-Multilingual-Auto-Tagger

2. Edit compose.yaml

At minimum, change:

• The auto-tagger database path
• The Ollama model storage path
• PUBLIC_BASE_URL
• The timezone, user ID and group ID when needed

For example:

PUBLIC_BASE_URL: http://192.168.1.100:8787

Replace the example address with the IP address of the Docker host.

3. Start the auto-tagger and Ollama

docker compose up -d --build

Pull the configured Ollama model:

docker exec -it arabic-sort-ollama ollama pull qwen3:8b

4. Update the Picard container

Use compose.picard-example.yaml as a reference.

The Picard container must include:

environment:
  ARABIC_SORT_API_URL: http://arabic-sort:8787
  ARABIC_SORT_PUBLIC_URL: http://YOUR-SERVER-IP:8787

It must also be attached to the same Docker network:

networks:
  - arabic-sort-network

At the bottom of the Picard Compose file:

networks:
  arabic-sort-network:
    external: true
    name: arabic-sort-network

For Arabic text, the jlesage Picard container should also install an Arabic-capable font:

INSTALL_PACKAGES: font-noto-arabic

5. Install the Picard plugin

The installable plugin archive is located at:

picard-plugin/arabic_tagging.zip

the zip plugin must be copied to config directory of picard-jlisage "import" folder After that open Picard and go to:

Options → Plugins → Install Plugin

Select arabic_tagging.zip, enable the plugin, and restart Picard.

The editable source is available at:

picard-plugin/plugin_source/arabic_tagging/__init__.py

Important notes

• LLM-generated titles may be incorrect. this depends on the model and complexity of titles also i found out after saving corrections for more albums e.g 20, the workflow became much more usable because previously verified titles were reused from the database instead of being generated again
• Review every suggested title before saving metadata.
• Do not expose Ollama or the auto-tagger service directly to the public internet without proper authentication and security.

Current limitations

• This pligin API for picard v2 i know picard v3 is in pre-release but this likely wouldnt work with picard v3
• The plugin is currently designed around the jlesage Docker version of Picard. (should work on the official picard but not tested as i use jliesage
• The project has mainly been tested with Arabic.
• The code needs review and restructuring by an experienced Python developer.
• Installation is still more technical than I would like.

Ideas for future development

• Add a button that writes the selected title directly into Picard after generation
• Support optional APIs such as OpenAI or Claude alongside local Ollama models
• Add proper language selection instead of assuming Arabic, maybe in "mode" or "language profile"
• Improve support for transliteration, translation and known localized titles
• Separate metadata parsing from LLM and database service calls
• Package the project as a more standard Picard plugin
• Improve Docker setup and documentation
• Rename the remaining Arabic-specific internal variables
• Explore whether the project could eventually be accepted into the official Picard plugin ecosystem

Development help wanted

I am not a professional developer or programmer.

I built this because I could not find an existing solution for my multilingual music library, and the current version is working well in my own setup. A large part of it was created with AI-assisted coding, so I would really appreciate help from experienced developers who can review it, clean it up and make it more maintainable.

I am especially looking for someone interested in:

• Reviewing the current architecture
• Identifying security or reliability problems
• Refactoring the Python code
• Improving the Picard integration
• Defining a realistic first development milestone
• Helping turn this into a proper community-maintained project

Issues, suggestions and pull requests are welcome.

Contributing

Before making a major change, please open an issue describing what you would like to work on.

For smaller fixes, documentation improvements and code cleanup, feel free to open a pull request directly.

Please keep changes focused and explain:

• What was changed
• Why it was changed
• How it was tested
• Whether it affects existing configuration or stored database entries

License

This project is licensed under the MIT License.

Acknowledgements

• MusicBrainz Picard
• The jlesage MusicBrainz Picard Docker image
• Ollama
• The open-source models used through Ollama

Disclaimer

I do not present this as a polished app and as mentioned earlier this was mostly vibe coded thus looking for devs to improve it and take AI slop out of it as i am not a programmer