Distinguishing infant cry from adult voice based on spectrum analysis

This paper proposes a method for distinguishing infant cry from adult voice based on spectrum analysis. The dominant components of spectrum of speech signal are used as features to discriminate infant cry from adult voices. The experimental results show that, under the same experimental conditions, the proposed method can completely reach real-time with discrimination accuracy of 98.82%, whereas the previous method fails to do so (about 4.5 s delays) with discrimination accuracy of 99.16%. That is, the proposed method can be run faster with high accuracy and therefore more effective for implementation on the micro-embedded system.