In this paper, we introduce a new multi-bit watermarking method for speech signals in the joint time-frequency domain. For this purpose, we compute the Wigner distribution of the speech signal and embed the watermark in a subset of time-frequency cells. The problem of synthesizing the signal from the watermarked Wigner distribution is discussed and a simple algorithm to reconstruct the speech signal from its Wigner distribution is proposed. The performance of the proposed watermarking algorithm is illustrated for sample attacks.