Training datasets

[soft swallow]

Most likely

But, if sample A is longer than sample B, sample A will have more impact on the final model

[soft swallow]

mmmmm... interesting

have you trained both with roughly the same amount of epochs orrr

I ask that bcuz artifacts can occur because of lack of training

Then it's most likely sample B bringing down the quality overall... hm

[digital nebula]

inconsistent audio is likely the cause of it. No matter how good your A is, if you add poor data on B, it affects the overall output, since rvc is averaging them out

Just the same as, combining dry vocals and wet vocals can ruin the quality