Abstract: A compressed speech system comprising a residual encoder for transmitting or storing a digitized speech signal having one bit of resolution. An original speech waveform may be reconstituted from the digitized signal to produce synthesized speech without the use of a vocal tract analog. The digitized signal is produced by generating a remainder or error signal between the original input signal and its predicted value. A predictor comprises an analog delay line in the form of a series of sample and hold modules, the outputs of which feed a corresponding series of correlation multipliers. The summed outputs of the correlation multipliers are an estimate of the next value of the input signal. This estimate is then fed back to the delay line input, forming a recursive filter which is continuously tuned to match the correlation statistics of the input signal.