A continuously varying electrical signal representing an original musical sound is divided at a selected mid-frequency of about 800 Hertz to provide upper and lower frequency signal bands which are then frequency-modulated with separate modulation signals that differ in frequency or amplitude or both, in order to produce a vibrato or tremulant effect and thereby enrich the harmonic content of the musical sound. In order to inject the vibrato or tremulant into each signal band, the modulation process utilizes a pulse sampling and delay circuit which minimizes clock pulse noise by selecting only the central part of each delayed pulse, and which also senses the original amplitude envelope of the unmodulated wave in order to augment the envelope that is recreated from the delayed pulses after they are recovered. The resulting modulated signals are recombined, producing a cross-modulation which adds further harmonic components, and the composite signal is again divided at a frequency of about 200 Hertz into high and low frequency output signals which are supplied to respective high and low frequency stationary loudspeakers.