Abstract: In order to reduce the noise introduced to speech signal by a laser Doppler vibrometer during the measurement of sound, the method of deep recurrent neural network(DRNN) speech signal denoising was adopted. The speech signal collected from laser Doppler measurement system was denoised. By using the deep recurrent neural network with 1 layer~3 layers and 1024 neurons per layer, the speech signals with signal-to-noise ratio from -6dB to 6dB were processed. After theoretical analysis and experimental verification, the results show that, as the number of layers increases, the quality of speech signals has risen to 8dB~12dB in many evaluation indexes. DRNN can effectively denoise the speech signals collected by laser Doppler acoustic system. The research is of practical significance for improving the quality of speech signals.