| 主权项 |
1. A method comprising:
detecting, by a phrase spotter running on a computing device, an utterance of a target phrase within an acoustic input signal, the detecting comprising applying one or more phrase spotting algorithms to the acoustic input signal; determining, by the phrase spotter, a first estimate of cumulative signal and noise energy for the detected utterance in the acoustic input signal, the first estimate being determined with respect to a first time period spanning a start time and an end time of the detected utterance; determining, by the phrase spotter, a second estimate of noise energy in the acoustic input signal, the second estimate being determined with respect to a second time period that precedes or follows the first time period; calculating, by the phrase spotter, a signal-to-noise ratio (SNR) for the detected utterance based on the first estimate and the second estimate; if the SNR is below an SNR threshold, rejecting, by the phrase spotter, the detected utterance as being an incorrect spot of the target phrase; and if the SNR is not below the SNR threshold:
accepting, by the phrase spotter, the detected utterance as being a correct spot of the target phrase; andcausing the phrase spotter or another speech recognizer to identify and process a verbal command spoken after the detected utterance. |
