Today, Google released Gemini 3.5 Live Translate, its latest audio model for live speech-to-speech translation. This isn't an incremental update to an existing feature. It's a purpose-built model that treats translation as a streaming audio problem, not a text pipeline with voice tacked on at the end.
Twenty years ago, translation at Google began as one of its pioneering machine learning experiments. That experiment has grown to over a trillion words being translated for billions of users across Google products every month. This release is where that trajectory meets real-time voice AI.
As someone who covers this beat daily, this one is worth watching closely. The combination of low latency, voice preservation, and an open developer API puts live translation in the hands of builders in a way it hasn't been before.
How Gemini 3.5 Live Translate Actually Works
Most voice translation systems today operate on a turn-by-turn model. You speak, you stop, the system processes, then outputs. Unlike those systems that "wait for the speaker to finish speaking before responding," Gemini 3.5 Live Translate "generates speech continuously, balancing the trade-off between waiting for context to improve quality and translating immediately to stay in sync with the speaker."
Older translation systems often wait for a full sentence because early words can be misleading until later words reveal tense, intent, or context. Gemini 3.5 instead runs streaming translation, where the model listens, interprets partial meaning, predicts what can safely be translated, and keeps updating as new speech arrives.
Gemini 3.5 Live Translate is a single audio model (gemini-3.5-live-translate-preview), not a chat assistant. Developers can configure it via the Live API using targetLanguageCode and echoTargetLanguage, with audio-only input and output: 16kHz in, 24kHz out. That's a clean, minimal interface for a model doing something technically demanding.
Key Technical Highlights
The model automatically detects 70+ languages and generates smooth, natural-sounding translated speech that preserves the speakers' intonation, pacing, and pitch.
- The translated output stays a few seconds behind the speaker throughout a session.
- The model handles multilingual inputs without the need to manually configure settings.
- Its noise robustness ensures applications can handle loud, unpredictable environments.
- Audio I/O specs: 16kHz input, 24kHz output, audio-only pipeline.
- All generated audio carries an imperceptible SynthID watermark for detectability.
The SynthID watermark is worth noting. Google said all audio generated by Gemini 3.5 Live Translate is embedded with SynthID, an imperceptible watermark designed to identify AI-generated content, integrated directly into the audio output.
