Live Voice Translations: Solving language barriers with seamless calls

Imagine a traveller landing in Bangkok who opens the Grab app and books a ride. The driver arrives two minutes early and calls to confirm the pickup point, but neither side speaks the other’s language. The call is confusing, adding friction to what should have been a seamless pickup. 

Southeast Asia is a linguistically diverse region, making situations like this a common occurence. Across our markets  travellers and driver-partners make more than 10 million voice calls every month, and many of those are between people who do not share a common language.

We saw an opportunity to make those interactions more seamless.

The live voice translation challenge

In recent years, text translation has become fast, accurate and widely accessible. But live voice translation is a very different challenge, where low latency and natural-sounding speech are just as important as translation accuracy.

At Grab, we’ve tested live voice translations with different models. Typical challenges we encounter are high latency, limited coverage for Southeast Asian languages, and translations that often sound flat or unnatural.

Testing Gemini 3.5 Live Translate

Our latest trials have been with Gemini 3.5 Live Translate, a speech-to-speech translation feature that can be integrated into in-app voice calls. The model is promising, as it provides low-latency, two-way translation in over 70 languages and is better at preserving speakers’ natural rhythm, pitch, and intonation. This is necessary for drivers and passengers  to understand each other in real-life situations.

The feature also auto-detects multiple languages while ensuring low latency. That’s helpful in our region with its diverse heritage, where everyday conversation can involve a mix of languages, such as Malay, English, and Chinese.