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

It's not really style transfer, but for a new speaker model, you just need to train each speaker with a dataset of 25 hours audio with time matched accurate transcriptions.

In the case of David Attenborough, I'm pretty sure that amount of data is available with subtitled BBC content.

Oh I doubt that will be necessary. I bet you could find a useful "feature vector" by repeating the process on about 10-20 new voices - then extract a set of words (let's pessimistically say 100) that activate these features the most - then have a person retrain with 5 examples of each of these 100 words (an hour, at the most).

I honestly wouldn't be surprised if you could find a really feature rich sentence (say 5 words long) that you could use to crack pretty much all voice-activated biometric password systems.

Is voice-activated password a thing? I mean, I only saw that in sci-fi movies and never in real life.

Yes, voice biometrics are used in telephone banking for instance

yep, I'm sure there would be a lot of similarity between models that could be used to reduce training time

Adobe VoCo does something like that. Demo: https://www.youtube.com/watch?v=I3l4XLZ59iw

It was pointed out the last time I posted this that the demo is "fake" at least in as much as the inserted words aren't being synthesized but rather copy-pasted with existing intonation intact.

That's pretty cool; any info on a release date?

My understanding is that while a number of people have tried, we currently have no particularly viable approaches for 'style transfer' analogues on sequence data, including text and sound/speech. Samples from past work are pretty bad. But what you could possibly do is roundtrip audio to text and then text (with conditioning like speaker metadata) to speech, both of which are doable. But directly transforming, not yet.

You mean, like they did pre-2000 [1]? Don't know how good this would sound today, but nevertheless interesting.

[1] http://www.washingtonpost.com/wp-srv/national/dotmil/arkin02...

I was wondering the same thing! I think, in the future this might be a problem for national security

Comedy option: Gilbert Gottfried.

Or Stephen Fry

Or Bill Kurtis? :)