Underdog shows off a 32MB speech model in new demo

Despite its small size, Underdog's new model can generate both speech and music.

<p>Despite its small size, Underdog's new model can generate both speech and music.</p>