warm tangle
Hi,
I am making a dataset in which not many samples exist of the single and even some samples are just very distorted due to the amount of vocal effects used on their production.
My dataset is roughly 6 minutes - I am wondering if I can use a dataset clip and add the copy to the pool to "extend" the dataset to 10 minutes? Or is it better to just use the 6 minute dataset?
harsh imp
It's better to just use the 6 mins, since you're just adding the same data/audio after it won't make a difference
You can try looking for other audios where the VA is speaking and use that instead to add more
warm tangle
Okay so in the training then is it better to train it at like 800 - 1000 epochs and make sure to cache? @still jolt
harsh imp
Yup just train it as is, you can train it with lower epochs if you cleaned it well enough