This paper deals with a new approach for perceptual audio encoding, based on the Empirical Mode Decomposition (EMD). The audio signal is decomposed adaptively into intrinsic oscillatory components by EMD called Intrinsic Mode Functions (IMFs), which can be fully described by their extrema. These extrema are encoded after an appropriate thresholding scheme controlled by a psycho-acoustic model. The decoder recovers the original signal after IMFs reconstruction by means of spline interpolation and their summation. The proposed approach is applied to different audio signals and results are compared to wavelets and to MPEG1-layer3 (MP3)approaches. Relying on exhaustive simulations, the obtained results show that the proposed compression scheme performs better than the MP3 and the wavelet approach in terms of bit rate and audio quality.