enquiry about configuration

[ZhouProject]

Hi @domkirke

I've been experimenting with various configurations and datasets but have yet to achieve satisfactory results with the MSPrior model.

I'm truly impressed by the MusicNet model with prior method you've successfully. Your results are quite remarkable.

Would you be willing to share some insights into your training process? I'm particularly interested in the configurations you opted for, such as the RAVE Architecture, any specific Regularization techniques for v2, and the MSPrior settings. Additionally, I'm curious about your dataset preprocessing steps—did you implement any unique methods, such as removing silent sections?

Your guidance would be incredibly valuable, and I greatly appreciate your help!

Many thanks,

Zhou