Among the long list of new features presented during Google I/O 2025 by the Californian company, real-time translation on Meet stands out. This is a significant addition that expands the app’s potential for virtual meetings, especially since the service has proven to be effective right from the start. During the keynote speech at the developer event, Big G showed a version of Meet that translates the sentences of two speakers, one in English and one in Spanish.
A surprising new feature
The result was surprising in terms of accuracy and reaction time between the words spoken by the speakers and their translation into the recipient’s language. However, it should be noted that the conditions for official presentations are ideal and prepared well in advance. Precisely for this reason, it is the actual functioning that leaves a positive impression, considering the advantages a service like this can offer.
Digital real-time translation breaks down one of the last language barriers, but the feature developed by Google also shows a great deal of realism. This is due to the short time lag between speech and translation but also because the reproduced voice is not robotic and impersonal, as it replicates the cadence, tone and emotions of the original voice. These details are crucial in attempting to credibly imitate the manner of expression of someone speaking in a language other than the one reaching the ears of the listener.
The role of AudioLM
The functionality is obviously based on Gemini, Google’s AI, although a crucial role is played by AudioLM, Google DeepMind’s audio-to-audio model, which skips the text transcription step in order to perform the translation. This system results in a minimal delay between the speaker’s spoken language and its translation, equivalent to the few seconds that professional interpreters need to perform their work.
Those who wish to do so can still listen to the conversation in the original voice, a useful option for improving foreign language learning. On a technical level, to activate live translation, simply start a call, select Voice Translation from the Meeting section and then indicate the languages you are interested in. To get results, translation must be enabled for everyone, with users involved in the dialogue having to consent to the translation of their voice using generative artificial intelligence.
After English-Spanish (bidirectional) support, Google has announced real-time translation on Meet between English and Italian (a surprise, given the more limited global spread of Italian compared to other languages). The next two languages available will be German and Portuguese, with the aim of reaching around ten languages by 2026.
Beyond the initial beta, real-time voice translation on Meet will be available in the coming weeks for Google AI Pro subscribers, which costs €21.99 per month after a free 30-day trial, and AI Ultra, which costs €249.99 per month. The higher cost of the premium subscription is determined by the ability to take advantage of Gemini 2.5 Pro, the most powerful version of Big G’s AI capable of tackling complex tasks such as programming, image and video production from text with the new Imagen 4 and Veo 3 models. In addition, the Ultra plan includes Project Mariner, an AI-based search system capable of automating up to 10 tasks simultaneously.
