Techniques for model training for voice features

A computing device, comprising:

• a processor;
• and
• a memory storing computer-readable instructions that, upon execution by the processor, configure the computing device to:
• maintain a language recognition model associated with a reference language, the language recognition model configured to accept a first utterance in the reference language as input and provide as output a feature corresponding to a set of device instructions that, when executed by a user device, cause the feature to be performed by the user device;
• obtain target language data of a target language, the target language data comprising a second utterance in the target language and an annotation associated with the second utterance, the annotation identifying the second utterance as being related to a particular feature that corresponds to a particular set of device instructions;
• generate training data for the language recognition model based at least in part by machine translating the target language data to the reference language;
• update the language recognition model utilizing the training data and a supervised learning algorithm;
• receive, from a particular user device, a subsequent utterance;
• and
• send device instructions to execute a corresponding feature at the particular user device based at least in part on a providing the subsequent utterance to the updated language recognition model as input, and receiving from the updated language recognition model output identifying the corresponding feature.

A computer-readable storage medium having stored thereon computer-executable instructions that, when executed by a processor, cause the processor to perform operations comprising:

• maintaining a language recognition model associated with a reference language, the language recognition model configured to accept a first utterance in the reference language as input and provide as output a feature corresponding to a set of device instructions that, when executed by a user device, cause the feature to be performed by the user device;
• obtaining target language data of a target language, the target language data comprising a second utterance in the target language and an annotation associated with the second utterance, the annotation identifying the second utterance as corresponding to a particular feature that is associated with a particular set of device instructions;
• generating training data for the language recognition model based at least in part by machine translating the target language data to the reference language;
• updating the language recognition model utilizing the training data;
• receiving, from a particular user device, a subsequent utterance;
• and
• sending device instructions to execute a corresponding feature at the particular user device based at least in part on a providing the subsequent utterance to the updated language recognition model as input, and receiving from the updated language recognition model output identifying the corresponding feature.