Google has up to date its Voice Search fashions to be powered by Speech-to-Retrieval (S2R). Google stated this permits it to “will get solutions straight out of your spoken question with out having to transform it to textual content first, leading to a sooner, extra dependable seek for everybody.”
Google initially used a voice search resolution named computerized speech recognition (ASR) to show the voice enter right into a textual content question, after which looked for paperwork matching that textual content question. Google stated this was “a problem with this cascade modeling method is that any slight errors within the speech recognition section can considerably alter the which means of the question, producing the improper outcomes.”
Speech-to-Retrieval (S2R) solved this situation. Google stated, “At its core, S2R is a expertise that instantly interprets and retrieves info from a spoken question with out the intermediate, and probably flawed, step of getting to create an ideal textual content transcript. It represents a elementary architectural and philosophical shift in how machines course of human speech.”
This was posted on the Google Analysis weblog however it’s getting used now, within the real-world. Google wrote, “The transfer to S2R-powered voice search isn’t a theoretical train; it’s a stay actuality. In a detailed collaboration between Google Analysis and Search, these superior fashions at the moment are serving customers in a number of languages, delivering a big leap in accuracy past typical cascade techniques.”
Hat tip to Gagan:
🆕 Big replace for Voice Search -> now its powered by Speech-to-Retrieval engine and this new course of do not convert speech to a textual content transcript & then do an online search quite this new approach makes use of an audio encoder for changing sound into audio embeddings which then is used to… https://t.co/iv2q4Kp0Qt pic.twitter.com/bCGwIfKNEh
— Gagan Ghotra (@gaganghotra_) October 8, 2025
Discussion board dialogue at X.
