Dynamic energy tracking for responsive voicing

A computer-oriented voicing decision is described which relies on the correspondence of low frequency energy content and voicing. This voiced /unvoiced decision compares the low frequency energy associated with successive, fixed duration speech intervals. The key to the decision process is the establishment of an adaptive energy reference level which enables the energy-based voicing decision to function independently of changes in overall speech volume. This energy reference level must have sufficient memory to permit inter-word comparisions, but must quickly adapt to rapid volume inflections such as those associated with weak, trailing voiced syllables.