Voice enhancement is of great significance to the development of intlligent on-board system and the future automobile industry.In order to solve the problem of driver voice noise pollution in the process of car driving,a least squares generation adversarial network model based on the efficient channel attention mechanism is proposed.Firstly, the attention mechanism is introduced in the generative network model to automatically select the one-dimensional convolution kernel size to generate the channel weight,which brings obvious performance gain,and then uses the least squares loss function to replace the Sigmoid cross-entropy los function to make the convergence rate faster and avoid the problem of gradient disappearance.Finally,the speech enhancement is realized.Experiments show that the proposed method has a good improvement in both quality and clarity over the baseline method,The PESQ index average increased by 3.79%,the STOI index average increased by 4.76%,so it is more suitable for practical applications.