Fig. 9From: A depthwise separable convolutional neural network for keyword spotting on an embedded systemContinuous audio stream test results. Hit rate and false alarm rates are shown as a function of SNR and detection threshold in a matched noise conditions (noise set 1) and b mismatched noise conditions (noise set 2). The trade-off between the hit rate and the false alarm rate can be controlled by selection of the detection thresholdBack to article page