Abstract: By studing the echo cancellation and reverberation cancellation methods of voice air conditioner in complex acoustic environment, so as to solve the problems of low recognition rate and poor robustness of far-field voice air conditioner. The frequency domain block adaptive filtering method is used for echo cancellation, the convergence speed is optimized for the large reverberation environment, and a multi-channel linear prediction method based on time-frequency mask is designed to adapt to noise reduction. Firstly, the deep learning technology is used to construct the signal time-frequency mask model, and then the multi-channel linear prediction adaptive late reverberation suppression technology is used to obtain the time-frequency mask information in combination with the mask model to suppress the reverberation components in the signal quickly and effectively. Finally, the spatial filtering technology based on transient data and eigenvalue decomposition is used in combination with the mask information, The multi-dimensional spatial signal guidance vector and noise statistics are accurately estimated, and the spatial filter is designed to effectively suppress the noise interference and improve the recognition robustness.

Key words: Frequency domain block adaptive, Echo cancellation, Multichannel reverberation suppression technology, Reverberation elimination, Noise suppression, Air conditioning application