Abstract: We propose a new perspective on speech
communication using blind source separation. The original speech is
mixed with key signals which consist of the mixing matrix, chaotic
signals and a random noise. However, parts of the keys (the mixing
matrix and the random noise) are not necessary in decryption. In
practice implement, one can encrypt the speech by changing the noise
signal every time. Hence, the present scheme obtains the advantages
of a One Time Pad encryption while avoiding its drawbacks in key
exchange. It is demonstrated that the proposed scheme is immune
against traditional attacks.