Announcing the update via a blog post, Google is attempting to make its Assistant more user-friendly as it tries to better recognise speech and pronounce unique names than it did before. Google announced that users can now teach the virtual assistant to enunciate and recognise unique names. The company said that the Assistant will listen to the users’ pronunciation and remember it, without recording the voice of the user. Currently the feature is available only in English but Google plans to expand to include other languages soon.
Another important feature introduced with the update is the ability to recognise the speech of a user and the context of the conversation in an easier manner. For this, Google has rebuilt Assistant’s Natural Language Understanding (NLU) models so it can understand context while also improving its “reference resolution.” The search giant said that this update uses machine learning technology powered by BERT. The company claims that due to this technology, Google Assistant can now respond with nearly 100 percent accuracy to commands for alarms and timers. Also, it is planning to bring this feature to other devices than just Google’s smart speakers that it is currently available on.
Google also applied BERT to improve the quality of the conversation between the Assistant and the users. Google Assistant will now use the users’ previous interactions and understand what is being displayed on the screen to respond to follow-up questions aiding in a more natural conversation.