In this research work, a novel approach to emotion identification system is proposed for implementation in audio domain using human speech. In order to undertake the new approach, average relative bin frequency coefficients will be extracted from speech. In a noisy environment, audio data are not strictly aligned, thus getting proper noiseless signal is a challenge. Consequently, this affects the performance of emotion detection system. Due to these reasons, a newly proposed approach of Average Relative Bin Frequency technique in frequency domain will be implemented through audio data. Support vector machine with radial basis kernel will be used for the classification. Preliminary results showed an average of 86 accuracy for average relative frequency bin coefficients. © 2012 IEEE.