Google Launches Gemini 3.8 Flash TTS, and Why Codec-LM Tokens Make |mhm| Backchannels Possible
prdeepakbabu · x · 2026-09-24
Google launched Gemini 3.8 Flash TTS and Flash-Lite TTS, its most expressive audio models: custom voices in 100+ languages or 2,000+ presets, line-by-line delivery direction, and natural conversational cues like <laughs> and active-listening interjections like |mhm|, with hours-long consistent generation.
Developer Deepak Babu Piskala highlights the real technical story: a backchannel must be emitted without claiming the turn, and mel-spectrogram TTS has no token to represent that — discrete codec-LM TTS makes it representable. He points to Chapter 7 of his book Building Speech AI, a 320+ page practitioner's guide with runnable code, notebooks, and CLI scripts spanning speech representation, recognition (Whisper, wav2vec 2.0, Conformer), and production voice-system trade-offs.
Related event: Google Launches Gemini 3.8 Flash TTS with 30-Second Voice Cloning(27 posts)→
More from Multimodal
- Open-source full-duplex speech model caramel-1: 413M params, sounds human but can't hold a conversation — alexcovo_eth · 2026-11-12
- H3-Image touts 16MP generation and realtime inference via 2-step LoRA on ZeroGPU — linoy_tsaban · 2026-11-12
- MiniMax H3 Image renders 4K portrait in half the time Seedream 5.5 takes for 2K — heyronir · 2026-11-12
- Endless screensaver of self-assembling quantum cryostats, made with Gemini 4 Carbon — hapticdata · 2026-11-12
- Apple's ADM 3 Cloud image model ranks #42 in editing and #85 in text-to-image on Artificial Analysis arena — ArtificialAnlys · 2026-11-12
- Adobe Firefly debuts Story Builder tool for streamlined visual storytelling — LudovicCreator · 2026-11-12