US 12,033,641 B2
Voice shortcut detection with speaker verification
Rajeev Rikhye, Freemont, CA (US); Quan Wang, Hoboken, NJ (US); Yanzhang He, Palo Alto, CA (US); Qiao Liang, Redwood City, CA (US); and Ian C. McGraw, Menlo Park, CA (US)
Assigned to GOOGLE LLC, Mountain View, CA (US)
Filed by Google LLC, Mountain View, CA (US)
Filed on Jan. 30, 2023, as Appl. No. 18/103,324.
Application 18/103,324 is a continuation of application No. 17/233,253, filed on Apr. 16, 2021, granted, now 11,568,878.
Prior Publication US 2023/0169984 A1, Jun. 1, 2023
This patent is subject to a terminal disclaimer.
Int. Cl. G10L 17/24 (2013.01); G10L 15/26 (2006.01); G10L 17/06 (2013.01); G10L 21/028 (2013.01)
CPC G10L 17/24 (2013.01) [G10L 15/26 (2013.01); G10L 17/06 (2013.01); G10L 21/028 (2013.01)] 20 Claims
OG exemplary drawing
 
1. A method implemented by one or more processors, the method comprising:
receiving audio data that captures an utterance of a human speaker and that also captures one or more additional sounds that are not from the human speaker;
processing the audio data and a pre-generated speaker embedding that corresponds to the human speaker, using a speaker separation model, to generate separated audio data which separates the utterance of the human speaker from the one or more additional sounds that are not from the human speaker;
processing the separated audio data using a text independent speaker verification model to generate speaker verification output;
determining, based on comparing the speaker verification output to the pre-generated speaker embedding that corresponds to the human speaker, that the human speaker spoke the utterance;
processing, the audio data or the separated audio data, using an automatic speech recognition (“ASR”) model to generate a text representation of the utterance; and
in response to determining that the human speaker spoke the utterance:
causing a client device to perform one or more actions that are based on the text representation.