Ethical AI Voice, Directing Performance & Protecting Human Talent - Dan Price

Voice cloning built to keep human actors employed

BionicVO is an AI voice platform built to keep human actors working rather than replace them. Its founder, a 35-year veteran of audio advertising, argues the scarce input is performance, the judgement that makes a read sound written by the actor, while rivals compete on voice count and gimmicks. The bet is that a good and ethical clone keeps brands from jumping to a cheaper one.

Takeaways

• AI voice rivals compete on volume, not performance. Price says rivals launch dozens of voices and gimmicks like switching a woman's voice to French mid-sentence, which no commercial he has seen needs. He has directed 12,000 to 15,000 commercials and says performance is the hard part.
• BionicVO builds its ethics around actor consent. Price called actors before agents or the union. Actors list conflicts such as a brand or a category like hard alcohol, and anyone who opts out never reaches a shortlist. Auditions stay watermarked and unusable until payment.
• AI threatens the non-union 80% of voice work. Since the 2000 strike, 80% of voiceover work has gone to non-union voices, the SAG-AFTRA group told Price. He guessed three or four years before AI handles it, the union 18 months. BionicVO aims to recapture that work.
• The roster is capped so members get work. BionicVO will not host thousands of voices. Price says there are not 4,000 people great at voiceover and puts the number between 90 and 260. The platform has 50 actors with a couple of dozen waiting.
• Human judgement is what a computer cannot copy. Casting goes to the actor who sounds as though they wrote the copy and is tethered to it. Price predicts AI voices will sound fully human within fifteen years, but says that distinctness is the part that stays human.
• A creative founder learned the tech business. Price had no idea what he was getting into. He learned how investments work, filed SEC transactions, set up a Zendesk help center and trained AI on it, moved development from Pakistan to a four-person US team and hired full-time support.

How BionicVO clones a voice

1. An actor records a few minutes of audio.
2. Four engineers spend about seven hours on 30 seconds of cloned audio.
3. The actor hears the clone and signs off before performance work starts, reacting with "that's awfully close" or asking for something to be tried differently.
4. Direction then adjusts pacing and adds emotive notes to steer delivery.

Where cloning breaks

• A cloned voice that sounded like the actor going in came out with a British accent.