An automatic sleep respiratory cycle segmentation method in time-domain is introduced. Unlike the methods using spectral analysis that require high computational process such as Fourier transform, our approach uses a nonoverlapping window for simple computation in time domain, which is suitable for low power computing systems. Typical sleep breathing sound files over 6-hour length were used to evaluate our method, and the experimental results showed that our method demonstrated the accuracy over 95% and possible improvement with further investigation on parameter optimization for the steps by close examination of the signal characteristics.