- Google DeepMind released Gemini 3.5 Live Translate, an audio model for live speech-to-speech translation.
- It automatically detects 70+ languages and preserves the speaker’s intonation, pacing, and pitch.
- Unlike turn-by-turn systems, it produces continuous, natural-sounding translated speech.
- Google says it translates over a trillion words monthly across its products.
What Happened
Google DeepMind released Gemini 3.5 Live Translate, its latest audio model for live speech-to-speech translation, according to its official blog. The model “automatically detects 70+ languages and generates smooth, natural-sounding translated speech that preserves the speakers’ intonation, pacing and pitch.”
DeepMind framed the release as the next step in a 20-year arc that began with Google Translate as one of the company’s early machine-learning experiments.
Why It Matters
Live, voice-preserving translation moves machine translation from text utility toward real-time conversation. It also extends Google’s AI product push into a high-visibility consumer feature, against a backdrop of rapid frontier releases tracked in the GPT-5.6 and Claude model race.
Technical Details
The model handles more than 70 languages with automatic detection and, unlike turn-by-turn pipelines, generates continuous translated speech rather than waiting for a speaker to finish. By preserving intonation, pacing, and pitch, it aims to keep the speaker’s vocal character intact across languages. Google says its translation systems now process over a trillion words per month across its products.
Who’s Affected
Multilingual users, travelers, and businesses running cross-language communication are the direct beneficiaries. The feature also pressures dedicated translation apps, since it ships inside Google’s existing product surface.
What’s Next
DeepMind’s blog post describes the model’s capabilities; broad rollout details, latency figures, and on-device versus cloud processing were not fully specified. Independent testing across low-resource languages will determine how evenly the 70-plus-language claim holds.
