2 ms·
Maybe not yet, but I am sure there will be speaker cloning extension soon, like photo and painting style transfer. Edit: It would seem they only need 25 hours
by visarga 10y ago
Maybe not yet, but I am sure there will be speaker cloning extension soon, like photo and painting style transfer.
Edit: It would seem they only need 25 hours of audio to train the model, so it would probably be easy to find or make alternate datasets.
> We train Tacotron on an internal North American English dataset, which contains about 24.6 hours of speech data spoken by a professional female speaker