National Taiwan University Academia Sinica
Can we make a famous rap singer like Eminem sing whatever our favorite song? Singing style transfer attempts to make this possible, by replacing the vocal of a song from the source singer to the target singer. This paper presents a method that learns from unpaired data for singing style transfer using generative adversarial networks.
Figure 1.
System framework for singing style transfer.
We have uploaded the paper to arXiv, please click the following link to download. arxiv 1807.02254, 2018.
Here we show the results of our singing style transfer model (the current best one).
There are total 6 clips, where
3 of them are for
male-to-female and the other
3 are for
female-to-male.
All of the
results below are
inside test!
Testing audio clips may be available to the models during training since we extract the spectrograms with overlaps.
Here we transfer 6 vocal clips in DSD100 (the dataset for singing voice separation) to validate our model's capability of generalization.
There are total 6 clips, where
3 of them are for
male-to-female and the other
3 are for
female-to-male.
All of the
results below are
outside test!
(we use only vocal source tracks to be the input!)
Testing audio clips are never be seen by the models.
