Voice activity detection with voice/nonvoice classification using reliable fundamental frequency

In this paper, we introduce a voice activity detection (VAD) algorithm to perform voice/nonvoice (V/NV) classification using a fundamental frequency (F0) estimator called YIN. Although current speech recognition technology has achieved high performance, it is insufficient for some applications where high reliability is required, such as voice control of powered‐wheelchairs for handicapped persons. The proposed VAD, which rejects nonvoice input in preprocessing, is helpful for realizing a highly reliable system. Previous V/NV classification algorithms have generally adopted statistical analyses of F0, the zero‐crossing rate, and the energy of short‐time segments. A combination of these methods, a cepstrum‐based F0 extractor, has been proposed [S. Ahmadi and S. S. Andreas, IEEE Trans. SAP. 7, 333–339 (1999)]. The proposed V/NV classification adopts the ratio of a reliable fundamental frequency contour to the whole input interval. To evaluate the performance of our proposed method, we used 1360 voice commands and 736 noises in powered‐wheelchair control in a real environment. These results indicate that the recall rate is 97.4% when the lowest threshold is selected for noise classification with precision 97.3% in VAD. The proposed VAD, which rejects nonvoice input in preprocessing, can be helpful to realize a highly reliable system.