Gemini 3.8 TTS lands in the API: 2,000+ voices, 30-second cloning
Google's new speech stack designs voices from a text prompt and clones them from 30 seconds of audio; plus GitHub Copilot's Slack and Teams upgrade.

Copy markdown
Gemini 3.8 Flash TTS and Flash-Lite TTS go GA
Google's two text-to-speech models are live in the Gemini API and AI Studio, growing from 30 voices to 2,000+ across 100+ languages and dialects. Flash TTS targets studio-grade, long-form audio like audiobooks and podcasts; Flash-Lite is the cheap, fast tier for dubbing and real-time voice agents. Text input runs to 8K tokens, audio output to 64K, all watermarked with SynthID.
Build a voice agent this weekend: prompt-designed voices + 30-second cloning
You can now spin up a custom voice from a plain-language description, or replicate one from a 30-second sample with consent checks, then stage two-speaker scenes with laughs, sighs, and line-by-line direction. It drops into LiveKit, Pipecat, Vercel, and Agora, so an automated podcast, a talking NPC, or a multilingual dubbing pipeline is now a weekend build.
GitHub Copilot in Slack and Teams gets sticky model choice
The chat bot now ingests files, attachments, and message links as context, links its work bidirectionally to GitHub issues and PRs, and lets you switch models mid-thread and keep that choice. It's a public preview for Copilot Business and Enterprise, billed against existing entitlements.