What's this about?
Google's two speech models support prompt-designed characters, long-form delivery, two-speaker scripts and fine-grained acting directions.
What is voice replication?
A short sample is used to reproduce a person's vocal characteristics. Unlike selecting a stock voice, it can imitate identity, so consent and provenance controls are essential.
Why does this matter?
official DeepMind X post 401 likes; builder demo post 362 likes.
- Replication requires a matching verbal consent recording.
- SynthID and C2PA aim to make generated speech traceable.
- Vendor benchmarks do not measure every impersonation or abuse scenario.