AI summary by TwinMind
What if You Could Sing in Your Favorite Musician's Voice? | Holly Herndon | TED
TED
9m 24s
Holly+ AI Voice Introduction and Technical Overview
- Holly (artist) introduces Holly+, AI trained on her voice
- Holly+ can sing in multiple languages Holly doesn't speak
- Demonstrates "El Cant de la Sibilla" (traditional Catalan song arranged by Maria Arnal)
- Notes melismatic runs are difficult to hit
- Performs "Mack the Knife" by Brecht in German
- Performs Holly's own song "Frontier" with overlapping voices
Technical Process: Timbre Transfer and "Spawning"
- Holly+ uses timbre transfer process
- Maps timbre/sound quality of one sound onto another's performance
- Creates machine-learning model of Holly's voice
- Holly recorded wide variety of phrases across entire vocal range
- Current version reads notes from musical score
- Demonstrates Spanish performance: "Bésame Mucho" by Velázquez
- Introduces concept of "spawning"
- Teaching AI sonic properties of one sound to generate entirely new sound
- Allows Holly+ to create vocals Holly didn't sing
Spawning vs Sampling: Conceptual Framework
- Spawning as 21st-century corollary to musical sampling tradition
- Sampling: copy/remix existing recordings to create new work
- Spawning: perform as someone else based on trained information
- Reframes artist's past work as potential reanimation archive
- Raises questions about collective human archive reanimation
- Creates new ethical and intellectual property challenges requiring new legal frameworks
Identity Play and Open Access Philosophy
- Introduces "identity play" concept for intellectual property
- Makes Holly+ freely available online for public use
- Explores possibilities:
- Others touring as Holly with permission
- Holly in thousand different bands, multiple languages
- Acknowledges musicians traditionally protective of IP for good reason
- Seeks mutually beneficial approaches while maintaining approval rights for derivative works
Live Performance Demonstration
- Invites musician Pher to stage
- Pher performs original song "Murky"
- Two microphone setup:
- One microphone: Pher's natural voice
- Second microphone: live Holly+ voice (developed with Voctro Labs)
- Full live performance of "Murky" switching between both voices
