Fig. 10From: A depthwise separable convolutional neural network for keyword spotting on an embedded systemContinuous audio stream test results showing the mean hit rate and false alarm rate in the range 0−20 dB SNR. Results are shown for the test network and the baseline network in matched and mismatched noise conditions. The test network was quantized to 8-bit activations and weights while the baseline network used floating-point precisionBack to article page