2019 Volume 2019 Issue AGI-012 Pages 06-
In this paper, we present a way to distinguish human voice from animal voice. In deep learning-based solutions, we adopted one class classification called DOC and then two class classification called VoVNET to verify all kinds of animal voice and human voice collected on the Internet. We construct a device to react only human voice, starting the voice recognition engine. It has a superior potential when used in a interactive AI in terms of economy. In the result of our study, the accuracy reached about 95% in DOC and over 98% in VoVNET.
