The invention provides a method and system for training a far-field speech acoustic model. The method comprises the steps of mixing near-field speech training data and far-field speech training data,and generating mixed speech training data, wherein the far-field speech training data is obtained by performing data enhancement processing on the near-field speech training data; training a deep neural network by using the mixed speech training data, and generating a far-field recognition acoustic model. The method and system provided by the invention can avoid a problem that the recording for far-field speech data needs to speed a lot of time cost and economic cost in the prior art; and the method and system not only reduce the time and economic cost for acquiring the far-field speech data,but also improve the far-field speech recognition effect.