Research on the robustness method of speech recognition in low SNR environment

Aiming at the sharp drop problem of speech recognition rate under noisy environment, an algorithm combining the improved minimum variance distortionless response beamforming and the improved Wiener filter based on time-frequency sparsity of speech is proposed in this paper. The algorithm first utilizes the spatial information of the microphone array speech signals to enhance the speech signal in the target sound source direction and to suppress the noise interference from other directions by the improved minimum variance distortionless response beamformer based on time-frequency masking, then uses an improved Wiener filter to remove residual noise and improve speech intelligibility. The mel-frequency cepstrum coefficients are extracted from the enhanced signal as characteristic parameters to build a speech recognition system of hidden Markov model. The experimental results show that the method proposed in this paper can effectively improve the speech recognition rate under low SNR environment and has strong robustness.