Google ships Gemini 3.8 Flash TTS to GA with voice cloning from a 30-second sample
2026-09-22
Google released Gemini 3.8 Flash TTS and Flash-Lite TTS to general availability along with a new Gemini API Voices endpoint, offering a library of over 2,000 prebuilt and custom voices. The API's "Voice replication" feature can create a custom voice from a 30-second audio sample "with consent verification," per Google's changelog. Simon Willison built a demo playground using the API's open CORS policy the same day, noting a two-pelican demo clip cost 2.74 cents to generate.
Significance 3: General availability of a voice-cloning feature from a short audio sample is a real capability with meaningful misuse potential, though Google states it requires consent verification and this is an incremental TTS product release rather than a frontier capability shift.
Implications · machine-drafted, not owner judgment
Consent-gated voice replication lowers the bar for legitimate custom-voice products (accessibility, dubbing, personal agents), but the strength of Google's "consent verification" is untested in practice — a demo builder found the API usable for realistic multi-character audio within a day of release, which is also the bar an impersonation attempt would need to clear.
- Reports of voice-replication consent checks being bypassed or abused
- Adoption of the Voices endpoint in production personal-agent products
Sources
- primary Gemini API release notes retrieved 2026-09-24
- curator Gemini 3.8 TTS Playground retrieved 2026-09-24