For communication applications where secure speech signal transmissions are a key requirement, speech scrambler is taken into consideration. To prevent someone from listening in on private conversations without their knowledge, it can transform clear speech into a signal that is unintelligible. The proposed speech scrambling system involves using two types of frequency transformation techniques: multiwavelet transform and Arnold transform. The effectiveness of the scrambling algorithm was evaluated with the help of three different measurements: the peak signal to noise ratio (PSNR), the estimated time (ET), and the mean square error (MSE). According to the final findings, the outcome of the scrambled speech signal does not have any residual intelligibility, while the quality of the descrambled speech is extremely satisfactory and has a low MSE level.