Google Construe is one of the company’s most used products. It helps people construe one accent to another through typing, taking pics of text, and using speech-to-text technology. Now, the company’s ablution a new activity called Translatotron, which will offer direct speech-to-speech translations – after even using any text.

In a post on Google’s AI blog, the team behind the tool explained that instead of using speech-to-text and then text-to-speech to catechumen voice, it relied on a new model (which runs on a neural network) to advance the new system.

“Dubbed , this system avoids adding the task into abstracted stages, accouterment a few advantages over cascaded systems, including faster inference speed, artlessly alienated compounding errors amid acceptance and translation, making it aboveboard to retain the voice of the aboriginal apostle after translation, and better administration of words that do not need to be translated (e.g., names and proper nouns),” the Google analysis team wrote in the blog post.

Translatotron can also bottle the characteristics of the voice of the apostle when advice from one accent to another. This could be really useful to sound editors who dub movies and TV shows.

The advisers have accepted that translations from the new model are not as absolute as acceptable models, but they’re assured the accurateness of the new model will soon improve.

There are plenty of apps out there like iTranslate and SayHi that try to construe one accent to addition using voice. But they’re still not as smooth and error-free as one would like.

Considering this still a model (and theres not even a demo accessible yet), affairs are Google will take a while to apparatus the new system in consumer-grade solutions. I, for one, am attractive advanced to trying it out though.

Read next: This IBM affidavit affairs shows you how to create AI-powered chatbots