A new analog speech scrambling technique that can be looked upon as processing the analog signal in two dimensions of time and frequency is introduced. A parallel bank of time-variant filters and a dither random pulse amplitude signal, both generated from multinonlinear direct-sequence spread-spectrum codes, operate on the analog signal to destroy its intelligibility. At the receiver, a reversed sequence of time and frequency processing takes place; with appropriate synchronization through a preamplifier, the analog received signal is recovered, with a minimum amount of distortion. The building blocks are introduced, and a statistical analysis is conducted to evaluate the security of the scrambling system, in terms of the autocorrelation and cross covariance. The security of the hybrid system is compared to that of other one-dimensional scrambling schemes, and derivations thereof.>