Exploring AudioMuse‑AI: A Self-Hosted Sonic Analysis Service for Music

[Relocate0604 8]

Emby’s music experience has long left much to be desired, and frankly, it hasn’t improved significantly over the years. I’ve come to believe that a standalone client for music might be the way forward, but since that isn’t currently planned, it’s worth exploring what’s possible today.
 

AudioMuse‑AI (GitHub link) is a new self-hosted tool that performs sonic analysis of your music library. While this kind of functionality was historically associated with Plex, AudioMuse‑AI is breaking new ground. It’s still in beta, but already supports Jellyfin and — thanks to the similarities in their APIs — can partially work with Emby.

 

Here’s what it can do:

- Creates a database of sonic analysis for your music library.

- Lets users define their own sonic fingerprints based on listening habits.

- Automatically generates playlists based on mood or keywords, using Ollama or other services.

During my testing on Emby:

- I was able to connect to Emby, access my music library, and generate a complete sonic analysis. On my hardware, this took about 72 hours.

- Ollama integration had issues but Automatic playlists could be generated using the Gemni integration. I am sure i did some wrong configurations. I hope there is a plan to integrate Docker Model Runner to avoid such issues in the future.

-  I could not save the generated playlists to Emby. It seems this is where Emby and Jellyfin APIs are not compatible.

There are still many improvements and bug fixes needed (see issues and road map). Without dedicated Emby support, this service can’t yet be fully utilized in the Emby ecosystem.

It’s important to recognize that there’s only so much a single developer can do. Full Emby integration would not just require explicit use of the Emby API, but also a plugin that allows each user to create their own dynamic playlists directly within the interface.

I’m sharing this because AudioMuse‑AI is highly relevant and has the potential to significantly improve the music listening experience for self-hosted users. It’s an exciting step forward — and hopefully a glimpse of what a fully integrated music experience in Emby could look like.

[Luke 42077]

HI, sounds interesting.

[Relocate0604 8]

Support for Emby servers has been added in the last beta release:
https://github.com/NeptuneHub/AudioMuse-AI/blob/main/deployment/docker-compose-emby.yaml
I was able to process a library of 2500 items each about 30min long in about two days on a laptop processor. 
I recommend using RAM to hold files being processed to reduce unnecessary disk writes:

  tmpfs:  # Use tmpfs for temporary audio files instead of a volume
      - /app/temp_audio:rw

Also they are working an OpenAI API integration. what is cool about it is that it works with Docker Model Runner (DMR) to add a small LLM to the compose file. no need to get Ollama running.

The only thing missing is an Emby Plugin: creating an automatic playlist from a song or a prompt is incredibly nice to have.

[thekingpin 29]

Really cool you got it working. This plugin is also the gateway to bring "Sonic Analysis" to Emby, a feature otherwise exclusive to Plex Pass. 

 

Im actually testing the Audiomuse jellyfin plugin and using Symfonium as client, so far its been really promising. 

I would love to hope someone has the knowledge, time, and, interest to create an emby plugin. 

[dark_aegis 3]

Would love to see this become a plugin eventually. One of the last things I'm missing from my time with Plex, really, and it makes a huge difference in mixes and stuff. Symfonium hooks into it for Jellyfin and Plex, so I'm sure if we ever got the functionality in Emby it could be added.