Latest audio model for live speech-to-speech translation
Gemini 3.5 Live Translate is Google's latest audio model purpose-built for live speech-to-speech translation, designed for developers, enterprises, and everyday users who need fluid cross-language conversations. At its core, the model delivers near real-time translation across over 70 languages while preserving the speaker's natural intonation, pacing, and pitch. This approach transforms what was once a clunky, turn-by-turn experience into a seamless dialogue that feels as natural as speaking in a single language. Available through the Gemini Live API, Google AI Studio, Google Meet, and the Google Translate app on Android and iOS, it brings advanced AI translation to billions of users worldwide. The model represents a leap forward from previous systems by generating speech continuously rather than waiting for pauses, making multilingual interactions feel immediate and human.
Traditional speech translation systems suffer from awkward pauses and delays because they wait for a speaker to finish before processing and responding. This turn-by-turn approach disrupts the natural rhythm of conversation, making it difficult to have flowing discussions across languages. Gemini 3.5 Live Translate solves this pain point with its continuous generation capability, which balances the need for context with the desire for immediacy. The model stays just a few seconds behind the speaker throughout the session, delivering fluid audio without silent gaps. For professionals in multilingual meetings, travelers navigating foreign environments, or customer service scenarios, this elimination of latency is critical. It allows participants to react in real time, ask follow-ups, and maintain engagement without the frustration of waiting for translations.
One of the standout features is continuous speech generation, which distinguishes Gemini 3.5 Live Translate from earlier turn-by-turn models. Instead of requiring the speaker to finish a full sentence, the model processes audio as it streams and generates translated speech incrementally. This technical approach means listeners hear translations with only a slight delay, closely matching the original speaker's pace. Combined with auto-language detection across 70+ languages, users do not need to manually configure source or target languages—the model identifies them automatically. This is especially valuable in scenarios where multiple languages may be spoken in the same session, such as international conferences or mixed-language family calls. The result is a natural, conversational flow that respects the rhythm of human speech.
Noise robustness is another major capability explicitly designed for real-world environments. The model can handle loud, unpredictable settings such as busy streets, crowded lobbies, or noisy call centers without significant degradation in translation quality. This is achieved through advanced audio processing that filters background sounds while focusing on the speaker's voice. For example, a traveler using Google Translate's new listening mode on Android can hold their phone to their ear during a noisy guided tour and still receive clear, translated audio directly through the earpiece. Similarly, enterprise users in open-plan offices can rely on the model during important cross-language meetings. This resilience ensures that speech-to-speech translation remains reliable wherever conversations happen.
Publisher
A
admin
Launch Date2026-06-11
Pricingfree
Domain Authority
Ahrefs DR91/100
Traffic Insights
Month2026-05
Estimated Visits15M
MoM Visits Growth+52.1%(+5.1M)
Avg Time on Site58s