In this paper a new algorithm is presented for speech encryption. It includes dividing the speech signal into overlapped blocks then shuffling those blocks in the time domain. A second permutation is done for the coefficients of the block which generated from wavelet transform by using chaotic key based on Hénon map, and partially encrypting the shuffled online speech signal in a transform domain. The security for the proposed system will depend on different parameters, including secret block sizes, the amount for overlapping along the x-axes and y-axes, permutation key and dynamic encrypted key. By having a new level of security the eavesdropper has to choose the amount of overlapping correctly. Many online speech signal tests demonstrate the validity of the proposed algorithm. The results show that it provides secure approach to real-time speech encryption and at the same time gives high intelligibility for the recovered speech.