Sudden changes in loudness of the sound signal between successive television scenes are avoided by substituting faded transitions between the sound signals of successive scenes instead of simple abutment of sound signal sequences one behind the other. This is produced, in a system in which the sound signals are digitally encoded, by distributing digital sound signal samples individually or by pairs to two recording channels in alternation and providing the transitions from the sound signals of one scene to the sound signal of the next scene earlier in one channel than in the other, by about 10 milliseconds. During that time span, one signal is faded out and the other is faded in by digital filtering or multiplication. It is convenient to provide such a reciprocal fading operation for the sound signals between each picture field or at least between each picture frame. During the reciprocal fading intervals, the sound signal samples of each sequence may be filled in with interpolated sample values inserted at the same time that a sample value of the other sequence is provided in the other channel. The provision of continuous forming of interpolated values can be used for substitution of interpolated values for sample values of a signal that has dropped out or been subjected to errors by disturbances.