Google DeepMind: The Podcast

Me, myself and AI

4 snips
Feb 22, 2022
Ask episode
AI Snips
Chapters
Transcript
Episode notes
ANECDOTE

Hannah's AI-Generated Voice Surprise

  • The episode starts with a WaveNet AI-generated version of Hannah Fry's voice that mimics her so closely it surprises her.
  • This showcases early impact of DeepMind's technology in realistic voice synthesis.
INSIGHT

WaveNet Improves Natural Speech

  • Traditional text-to-speech involves stitching together prerecorded voice bits, causing robotic-sounding speech.
  • WaveNet directly models raw audio waveforms, producing more natural and fluid voices.
INSIGHT

WaveNet's Fine Tuning Breakthrough

  • Initially, WaveNet required around four hours of audio to model a voice, now only a few minutes suffice.
  • This is achieved through fine tuning by leveraging a large dataset from professional speakers.
Get the Snipd Podcast app to discover more snips from this episode
Get the app