How to clone your own voice with Gemini 3.8 TTS: Phil Schmid's hands-on guide

_philschmid · x · 2026-09-24

Phil Schmid published a hands-on guide for the newly available Gemini 3.8 Flash TTS and Flash-Lite TTS, which top Hume's Voice Design Benchmark and Voice Arena in 6 languages. Key tips: record two clips with the same mic and room at 24kHz mono, use -f alsa -i default on Linux (re-run after granting macOS mic permission), and read one of the 25 exact consent sentences. The guide shows both voice replication and creating a new voice from a single sentence, with a prompt you can hand to your agent.

Original post →

More from Multimodal

Multimodal channel →