Google plans to add a live transcription feature to its Google Translate app for Android at some point in the future. The feature will allow users to record audio in one language and have it rendered in another in real time. It’s still in the prototype stage, but Google gave a demonstration of the technology during a series of artificial intelligence demos at its San Francisco office.

Google told The Verge the need for an internet connection is because real-time multilingual transcription is more complex than translating text from one language to another. Further, the software won’t work with audio files initially. Instead, the transcription service will need to capture live audio through a smartphone’s microphone. However, Google did note that users could play recorded audio and capture it in that way with the transcription feature. The search giant explained how the feature works, saying it constantly evaluates whole sentences as the audio goes on. It then adds punctuation, corrects certain word choices depending on context and attempts to correct for things like accents and regional dialects. While that happens, users should receive an accurate approximation of what’s being said. Google expects the AI models behind the feature to improve significantly as time goes on.

Google says its transcription feature is constantly evaluating whole sentences as the audio is ongoing. From there, it’s adding punctuation, correcting certain word choices depending on the sentence’s context, and attempting to correct for things like accents and regional dialects. All the while, users should be getting what Google says is an accurate approximation of what’s being said, although the company expects the underlying AI models to improve significantly as time goes on.

There’s no concrete timeline for when Google plans to launch real-time transcription for Google Translate. The company is just saying it’s arriving at some point “in the future.”

At the moment, Google Translate already helps users converse in multiple languages and translates them in real-time. However, with this new feature, it will support longer-form conversations and speech. This means that if you’re trying to record a television show or a lecture that’s spoken in another language, then this upcoming feature will help you do that while transcribing and translating at the same time