The invention relates to a method for coding a variable speed audio frequency switching between adjacent high/low speed coding modes, which belongs to the field of audio coding and is particularly suitable for a multi-speed audio encoder. The method is technically characterized by comprising the following steps of: coding and decoding each frame of audio signals at a high speed, making coding input signals and decoding output signals in various coding modes at the speed pass through a sensing and weighting filter, calculating an average segmentation signal to noise ratio (SNR), and selecting a coding mode with the maximum sensing and weighting average segmentation SNR; selecting a coding mode with the maximum sensing and weighting average segmentation SNR of the coding input signals and the decoding output signals at a low coding speed close to the high speed; and finally, calculating average segmentation SNR of the coding input signals and the decoding output signals relative to the coding modes selected at the high speed and the low speed respectively, if the average segmentation SNR in the low-speed coding mode is greater than the average segmentation SNR in the high-speed coding mode, switching to the low-speed coding mode, otherwise, switching to the high-speed coding mode. By the method, each frame of audio signals are switched among various coding modes at adjacent high/low coding speeds according to distortion of the output signals relative to the input signals, so that high coding quality is kept, and the average coding speed of the audio signals is reduced simultaneously; therefore, the coding efficiency of the multi-speed audio encoder is improved.