MUMBAI, India, Oct. 31 -- Intellectual Property India has published a patent application (202547094976 A) filed by Google Llc, Mountain View, U.S.A., on Oct. 3, for 'mixture-of-expert conformer for streaming multilingual asr.'
Inventor(s) include Hu, Ke; Li, Bo; Sainath, Tara N; Zhang, Yu; and Beaufays, Francoise.
The application for the patent was published on Oct. 31, under issue no. 44/2025.
According to the abstract released by the Intellectual Property India: "A method (500) of a multilingual ASR model (200) includes receiving a sequence of acoustic frames (110) characterizing an utterance (106) of speech. At a plurality of output steps, the method further includes generating a first higher order feature representation (212) for an acoustic frame by a first encoder (210) that includes a first plurality of multi-head attention layers (300a); generating a second higher order feature representation (222) for a corresponding first higher order feature representation by a second encoder (220) that includes a second plurality of multi-head attention layers (300b); and generating, by a first decoder (240a), a first probability distribution (120a) over possible speech recognition hypotheses. A gating layer (410) of each respective MoE layer (400) configured to dynamically route an output (402) from a previous multi-head attention layer at each of the plurality of output steps to a respective pair of feed-forward expert networks (420)."
The patent application was internationally filed on Mar. 07, 2024, under International application No.PCT/US2024/018806.
Disclaimer: Curated by HT Syndication.
