Amplitude-bandwidth voice joint analysis of the robust detection

A novel method with amplitude-bandwidth analysis is proposed to overcome the difficulty of low SNR speech endpoint detection. By adaptively adjusting the noise power spectrum of each frame, the stochastic peak of noise spectrum is eliminated and the noise is restrained, then double decision thresholds are applied to detect the speech endpoint in the variance information of speech power spectrum and the noise one ratio. Simulation on computer shows that the new algorithm is more effective in distinguishing speech from noise and is more significant in improving the performance of automatic speech recognition system. It is proved to be robust under various noisy environments. The algorithm is of low computational complexity and is simple in real-time realization.