MUMBAI, India, Nov. 2 -- Intellectual Property India has published a patent application (202617093849 A) filed by Google LLC on August 03, 2026, for Speculative Asr Decoding To Reduce Overall Latency Of Speech Applications.
Inventors include Rosenberg, Andrew; Ramabhadran, Bhuvana; Yusuf, Bolaji; and Baskar, Murali Karthick.
The application for the patent was published on October 30, 2026, under issue no. 44/2026.
Abstract: A method (500) includes receiving a training utterance (106) including a sequence of audio frames (102) and a corresponding ground-truth transcription (320) and truncating a suffix portion from the sequence of audio frames to provide a prefix sequence of the audio frames. The method also includes processing, using a speech recognition model (200), the prefix sequence of audio frames to provide a partial speech recognition hypothesis (180a) and processing, using a prefix embedding multi- head attention layer (170), a sequence of prefix encodings (210) to generate a prefix prompt embedding (175). The method also includes processing, using a LM (160), the partial speech recognition hypothesis conditioned on the prefix prompt embedding to generate a speculated speech recognition (SSR) hypothesis (180b) and determining a training loss (325) based on the SSR hypothesis and the ground-truth transcription. The method also includes fine-tuning parameters of the prefix embedding multi-head attention layer.
Disclaimer: Curated by HT Syndication.
