Surface electromyography (EMG) is a bioelectrical signal that recognizes speech contents in a non-acoustic form. Activity detection is an important research direction in EMG research. However, in the low signal-to-noise ratio (SNR) environment, it is difficult for traditional methods to obtain accurate active signals. This paper proposes a new energy-based spectral subtraction backtracking (E-SSB) method to segment EMG active signal in the low SNR environment. Compared with traditional energy detection, the algorithm in this paper adds spectral subtraction (SS) to filter out the clutter, and raises a retrospective idea to improve the classification performance. The experiment results show the proposed activity detection method is more effective than other methods in the low SNR environment.