Digital audiowatermarking can be viewed as a communication system (cf. Figure 1) where the information $v_n$ is embedded imperceptibly into the digital audio signal $x_n$ through appropriate spectral shaping filter $H_j$ . The resulting watermark $t_n$ should be robust to standard signal manipulations (i.e. compression, A/D-D/A conversion, resampling,...) on one side, and to intentional pirat attacks on the other side. In particular, all-pass filter attacks introduce a nonminimum phase problem in the watermarking detection scheme of figure 1, which strongly degrades the performances of detection techniques based on mean square criteria. We propose in this paper a solution to the above mentioned problem, combining Wiener deconvolution and blind equalization based on a non quadratic criterion.