US 11,948,562 B1
Predictive feature analysis
William Evan Welbourne, Seattle, WA (US); Min Hao Chen, Seattle, WA (US); and Jennifer Liwen Chen, San Diego, CA (US)
Assigned to Amazon Technologies, Inc., Seattle, WA (US)
Filed by Amazon Technologies, Inc., Seattle, WA (US)
Filed on Dec. 11, 2019, as Appl. No. 16/710,756.
Int. Cl. G10L 15/22 (2006.01); G10L 15/26 (2006.01)
CPC G10L 15/22 (2013.01) [G10L 15/26 (2013.01); G10L 2015/223 (2013.01); G10L 2015/227 (2013.01); G10L 2015/228 (2013.01)] 18 Claims
OG exemplary drawing
 
1. A computer-implemented method comprising:
during a first time period:
receiving first data representing a first device name associated with a user profile, the first device name indicating a first user device;
determining a device name embedding using the first data, the device name embedding representing a machine readable representation of the first device name; and
storing, in a database, the device name embedding with the user profile;
during a second time period subsequent to the first time period:
receiving input data;
determining, using the input data, that a predicted user input is expected to be received at a future predicted time, the predicted user input corresponding to an intent to operate the first user device represented by the first device name;
determining that a duration of time is required to retrieve the device name embedding from the database causing a latency; and
retrieving, at a first time based on the duration of time and prior to the future predicted time, the device name embedding; and
during a third time period subsequent to the second time period:
receiving a user input;
determining the user input is associated with the user profile;
processing the user input and the device name embedding using a previously trained model to determine that the user input corresponds to the intent to operate the first user device; and
using the device name embedding to determine output responsive to the user input, the output causing operation of the first user device.