In order to improve the robustness and discrimination of the speech perceptual hashing function over the content based encrypted speech retrieval, increase the retrieval efficiency and accuracy, and realize the privacy and security of speech data. Biological hash has a good robustness, discrimination, and good protection for biometric template in the aspect of multimedia authentication as a feature extraction method. We propose a retrieval algorithm of encrypted speech based on biological hashing. Firstly, a two-dimensional Henon mapping encrypted algorithm is used to encrypt the speech and upload it to encrypted speech library of the cloud. Secondly, the speech signal is conducted with discrete wavelet transform (DWT) after pre-processing conducts to produce low frequency wavelet coefficients which are regarded as the wavelet Merlin matrix. Finally, the pseudo random matrix is generated by the Logistic mapping, and the pseudo random matrix is conducted with fast Fourier transformation (FFT) transform to produce pseudo random Fourier matrix. The wavelet Merlin matrix and pseudo random matrix are conducted with iterations and thresholds to produce the binary perceptual hashing sequence, and binary perceptual hashing sequence stored in the system hash index table of the cloud. When retrieving data, the Hamming distance is used to match and retrieval. Experimental results show that the hash function of the proposed algorithm has excellent robustness, discrimination and security, as well as the efficiency and accuracy of retrieval model is obviously improved.