Twenty Thousand Hertz

Deepfake Dallas: How AI learned to speak like me

Sep 2, 2020
Ask episode
AI Snips
Chapters
Transcript
Episode notes
ANECDOTE

Dallas Creates Own Deepfake Voice

  • Dallas Taylor created an audio deepfake of his own voice for the episode.
  • This exemplifies the ease and creativity in making voice deepfakes today.
INSIGHT

How AI Learns to Speak

  • AI learns to speak by mapping text to audio patterns.
  • Initial outputs are robotic but improve as training data accumulates.
INSIGHT

Training Efficient Deepfake Voices

  • Fine-tuning a generic model with just hours of voice data creates a convincing deepfake voice.
  • Data compression is used to speed training but reduces voice quality temporarily.
Get the Snipd Podcast app to discover more snips from this episode
Get the app