Abstract: Selection of training utterances may be carried out in a sample-efficient manner, and the selected training utterances may be annotated to provide improved training information to an ASR system. A computing device may receive, from an ASR system, one or more transcript-score pairs, wherein a transcript-score pair comprises a transcription associated with a voice query and at least one score associated with the transcription. The computing device may determine a likelihood of a word error associated with each transcription of the one or more transcript-score pairs. The computing device may determine, based on the likelihood of the word error, an effect on a word-error rate of the ASR system. The computing device may send at least one of the one or more transcript-score pairs with a threshold effect on the word-error rate of the ASR system to be annotated.

Abstract: A user device may detect speech and use “early exiting” when identifying a potential operational command in the detected speech. The implementation of early exiting may be based on a variable threshold, where variable sensitivity settings for the threshold may be used to control how quickly, and whether, an “early exit” or early prediction of an operational command will occur. An early exit threshold may be adjusted, for example, based on network conditions, to ensure optimal operational command determination from the audio.

Abstract: First audio data associated with a first portion of a voice query (e.g., an incomplete voice query) may be received (e.g., by a device or a server). A first transcript may be determined by a speech recognition engine and based on the first audio data. A plurality of predicted queries may be determined by applying a prediction process to the first transcript. A response for each of the plurality of predicted queries may be determined by processing the plurality of the predicted queries. Second audio data associated with a second portion of the voice query (e.g., a complete voice query) may be received. A second transcript may be determined by the speech recognition engine and based on the second audio data. Based on comparing the second transcript to one of the plurality of predicted queries, a response for the voice query may be returned.