We use an artificial neural network to recognize when a potentially noisy recording contains human speech, for voice-activated consumer electronics, machine-learning-assisted hearing aids, and more.