| CPC G10L 15/22 (2013.01) [G10L 15/187 (2013.01); G10L 15/25 (2013.01); G10L 2015/088 (2013.01); G10L 2015/223 (2013.01)] | 16 Claims |
|
1. A method comprising:
receiving, by a computing device, audio data, wherein the audio data comprises speech of a user and other audio;
isolating, by the computing device, speech of the user from the audio data;
identifying, by the computing device, at least one of a one or more vowel sound and a diphthong in the isolated speech of the user;
analyzing, by the computing device, the isolated speech of the user and generating a plurality of predictions, wherein each prediction of the plurality of predictions comprises at least one of a word and a phrase in the isolated speech of the user;
performing, by the computing device, a post processing on the plurality of predictions and generating a plurality of processed predictions based on the plurality of predictions, wherein performing the post processing on the plurality of predictions and generating a plurality of processed predictions based on the plurality of predictions comprises:
comparing, by the computing device, at least one word and phrase of the plurality of predictions to one or more relative words and phrases in a word-phrase graph;
calculating, by the computing device, a correction score for at least one prediction of the plurality of predictions and at least one relative word and phrase of the one or more relative words and phrases;
determining, by the computing device, whether the correction score exceeds a threshold correction score; and
generating, by the computing device, a processed prediction,
wherein the processed prediction comprises the at least one relative word and phrase used to calculate the correction score if the correction score exceeds the threshold correction score, and
wherein the processed prediction comprises the at least one prediction used to calculate the correction score if the correction score does not exceed the threshold correction score;
selecting, by the computing device, a first command in a command profile, the first command defining a user input comprising at least one of the vowel sound and the diphthong identified in the isolated speech of the user and the one or more processed predictions of the plurality of processed predictions, wherein a command comprises at least one user input mapped to at least one mapped task associated with a mapped device and to be executed by the computing device;
executing, by the computing device, the at least one mapped task of the first command; and
causing, by the computing device, at least one corresponding action at the mapped device associated with the one or more mapped tasks of the first command
wherein a correction score is calculated according to:
 wherein RW represents a weight of the relative word or phrase,
LW represents a weight of the word or phrase of the at least one prediction,
D represents a phonetic distance between the relative word and the word or phrase of the at least one prediction,
F represents a total number of times the word or phrase of the at least one prediction has been detected in previous iterations of post processing,
T represents a number of transitions made from the word or phrase of the at least one prediction to the relative word or phrase, and
H represents the threshold correction value.
|