Google Voice Search is now powered by their new Speech-to-Retrieval (S2R) engine. It was previously powered by automatic speech recognition (ASR), which is a different technology.

Here's what makes S2R better than ASR:

• ASR converts a verbal query into text, then uses the text query to search for relevant results

◦ This can lead to unhelpful search results if there's even a small error in recognizing the spoken phrase

• S2R skips the "verbal to text" step and directly retrieves information based on the spoken query

◦ How, exactly? No clue. This is why I'm not a Google engineer. The blog post goes into detail. And S2R isn't perfect either, it's just significantly more accurate than ASR.

The announcement was published on Google Research but it's not just a breakthrough in a lab, it's an actual implemented change to how voice search works as of now:

> "The move to S2R-powered voice search isn’t a theoretical exercise; it’s a live reality. In a close collaboration between Google Research and Search, these advanced models are now serving users in multiple languages, delivering a significant leap in accuracy beyond conventional cascade systems."