An improved speech endpoint detection based on adaptive sub-band selection spectral variance

Speech endpoint detection is an important component in different speech processing systems. The endpoint detection using basic spectral variance becomes difficult and inaccurate when speech signals are contaminated by high noise. An improved spectral variance method is presented in this paper. The noisy speech is firstly enhanced using spectral subtraction method. Then an adaptive sub-band selection spectral variance is used to detect the endpoints. It improves the discriminability between speech and noise so that it becomes easier to set thresholds. Finally, experiment results are given to show the effectiveness of the improved method outperforms basic spectral variance method. Furthermore, for low signal-to-noise ratio (SNR), the proposed method has better robustness for different types of noise.