DeepMind ships Gemini 3.8 Flash TTS with native multi-speaker overlap and laughter

kastnerkyle · x · 2026-09-24

Google DeepMind launched two new TTS models: Gemini 3.8 Flash TTS, which can design voices with distinct accents and characteristics, and Flash-Lite TTS, built for efficiency and scale with a production-ready voice library. Team member MosesOh4 highlights the key breakthrough—native multi-speaker speech where models can laugh together, backchannel and overlap naturally—paving the way from TTS and NotebookLM-style dialogs to real-time assistants. More audio generation models are teased.

Original post →

More from Multimodal

Multimodal channel →