Fig. 3
From: Room-localized speech activity detection in multi-microphone smart homes

Histograms of the five hand-crafted scalar features of Section 5.1, demonstrating their ability to discriminate room-inside vs. room-outside speech. Histograms are computed over the development set of the simulated dataset of Section 7.1, for the case of the smart home bedroom (see also Fig. 1). Upper row, left-to-right: energy-based feature (Section 5.1.1), coherence feature (Section 5.1.2), and envelope variance one (Section 5.1.3). Lower row, left-to-right: spectrogram texture smoothness feature (Section 5.1.4) and SRP-based one (Section 5.1.5)