Not Waveglow

Their other one they released

what exactly is nemo

i see it downloaded everytime i train talknet

It's just Nvidia's audio ML repo

They have all sorts of neat models here

GLOWTTS might've been it

I think nvidia is working on a new promising multispeaker model too, so waiting on that

Not Fastspeech?

Cant recall which one it was exactly

Ok Im dumb, heard multispeaker a lot but still dont fully grasp it, what is it?

Not fastspeech tho

Ability to produce distinct synthesizer outputs in different speaker styles (usually given some input to determine which style to adopt)

Ah kk

Trains on a bunch of different speakers tho

Instead of just 1, so can take a while

Ah right ok cool

Mellotron is multispeaker

Kinda so you can do voice style transfer?

AutoVC is quite different but looks kinda cool for voice style transfer

Oh wait my bad

It was flowtron I was thinking of

Takes a min of 2hrs training data for transfer learning, and more like 10hrs from scratch

Ahh ok, yea Flowtron looked good

Melotron is really intended for singing style transfer

Flowtron is what they used for their pitch demo synthesized voice

The nvidia keynote thing

Gets solid results but idk what the inference time is like