Voice channel

We introduce and offer a partial solution to the voice-channel problem. We are motivated by the problems in speech processing and recognition. The usual channel-capacity problem involves choosing appropriate signals to communicate over a fixed channel. In contrast, the voice-channel problem involves choosing appropriate channels or filters to modulate a random input signal from a known distribution. We want to characterize the set of distinguishable filters that maximize the information transmitted. The solution to this problem has tie-ins with achievable rates on Gaussian fading channels.