Is it possible to retain a model of the voice conversion?

[mazzzystar]

Is it possible to retain a model of the voice conversion?

[mazzzystar]

@HiroshiKawata
Current model actually don't learn anything because it's initialized totally randomly.
The training process is to repeatly optimize the source sound piece so that it's network output features meet our requirements, therefore the model parameter won't be updated. So there is not use to save model parameters.

However, your question reminds me of a recently paper(github implementation) which shows that image style transfer task can be learned by a linear model. So maybe it's reasonable to train a model like that.

[HiroshiKawata]

I see. Thanks for the explanation and the paper, @mazzzystar ! 😃