Özet:
In this manuscript, some voice activity detectors (VADs) in the literature were examined in terms of factors affecting their robustness under different acoustic noise conditions and in this context, the changes in detection accuracy rates according to changing noise conditions were tested. In this scope, the effect of situations such as whether the threshold value used in the decision phase in VAD methods is fixed or adaptive, the analysis window is short or long, the use of more than one feature vector together has been evaluated and analyzed comparatively. While three of the four different VAD detectors examined in this manuscript use feature vectors within the short-term analysis window while generating the decision result, one decides according to the measurement result of long-term spectral vectors. The VAD detectors in the article have been tested using the NOIZEUS noisy speech database. Thus, the performance of the analyzed VADs has been evaluated under different acoustic conditions using an extensive database that has already taken place in the literature. During the testing of the analyzed VADs, different input noise speech signals with environmental background noises between [15-0dB] such as restaurant, car, street, or station were tested. Tests were carried out using objective test measurement methods and the detection accuracy rate of each VAD method was measured. The results showed that each method gave different endurance performance in adverse environmental conditions.
