Performance is low on our custom data
habib1402 opened this issue · comments
Hi, we are currently training Spleeter using two classes: the machine class and the noise class. For training purposes, we are providing three classes as input: the machine class, the noise class, and a mix of both classes. Each class consists of 360 voices. However, despite conducting eight experiments, we have not observed any improvement in performance. We are seeking assistance to address this issue and improve the results.
The training is stopped.
Can you please tell me what these files are for and what I should do if I need to reproduce them in python, please