karaokenerds / python-audio-separator

Easy to use vocal separation from CLI or as a python package, using a variety of amazing models (primarily trained by @Anjok07 as part of UVR)
MIT License
300 stars 50 forks source link

Add a model #58

Closed shirounanashi closed 1 month ago

shirounanashi commented 3 months ago

Can you please add and host this model? It's a VR Arc model, and it's the best model I know for keeping and removing backing vocals. It works on UVR5 https://drive.google.com/file/d/1LSHlTvVt4FJxG8VGetueAuNlDc5w3Deu/view?usp=sharing

beveradb commented 3 months ago

Sorry, I'm not sure how this actually works with UVR - could you show me some screenshots / give me some instructions to explain how you currently use it with UVR? Is there a GUI feature where you can select your own model file or something?

If so, and if I'm able to get it working myself with UVR, I might be able to add support for that model - but at the moment, the output from audio-separator -l lists all supported models, and those are supported only because I'm able to extract the model configuration data from UVR itself (well, from the repositories on github which UVR uses to host config data).

If the model you've linked is legitimately supported by UVR, it must have configuration data hosted somewhere (potentially mapped using an MD5 hash of the model file) - in order to support it in audio-separator, I need to be able to find and access that model config data.

Hope that makes sense!

shirounanashi commented 3 months ago

I simply added the model to the folder "C:\Ultimate Vocal Remover\models\VR_Models". Regarding the config data, i found this in the file "C:\Ultimate Vocal Remover\models\VR_Models\model_data\model_data.json"

    "f6ea8473ff86017b5ebd586ccacf156b": {
        "vr_model_param": "4band_v2_sn",
        "primary_stem": "Instrumental",
        "is_karaoke": true
    },
shirounanashi commented 3 months ago

@beveradb Now I have another request! Can you add the BSRoformer models that were added in the latest UVR5 beta update?

shirounanashi commented 3 months ago

By the way, I managed to make the model I sent here work in Audio Separator, I just had to change the model check file and it worked without problems

beveradb commented 1 month ago

FYI @shirounanashi I've now released audio-separator version 0.17.1 with support for the BS-RoFormer and Mel-Band RoFormer models, e.g. model_mel_band_roformer_ep_3005_sdr_11.4360.ckpt

I highly recommend checking them out, the mel-roformer model performs better than pretty much everything else I've tried so far!

As for hosting/adding custom models, I'm not personally motivated to add support for that, but you're welcome to raise a pull request if you want to add that support yourself 😄