The translation itself is currently processed on Google’s AI-focused data-centers, because it takes a lot of processing power. Audio must be converted to text, translated into another language, and then turned back into speech and spoken to the listener
The last part of that process is traditionally done by putting together pre-recorded words or word fragments. However, DeepMind, Alphabet’s AI research lab, wrote in a blog post today that the AI research it used to generate human-sounding voices—a system called WaveNet—is now in Google Assistant. That means the voice speaking the translations will be generated in real time and thus more realistic, according to DeepMind. What’s unclear is how much of this processing will be done in the cloud and how much on the processor of the phone connected to the new earbuds.
The Google Pixel Buds cost $159 and provide 5 hours of battery life, and can be recharged from a battery pack in their carrying case.