Mixed distance measures for synthetic speech evaluation
Nathaniel Polish · 1991
A set of speech quality distance measures has been developed. These measures are called mixed distance measures because they apply different criteria to different classes of voicing. The results of the measures are combined in what might be termed a knowledge-based approach. These measures operate in both time and frequency domains and exploit particular characteristics of the class of voicing (voiced, fricative, stop) that is examined by the measure. These measures provide a method for speech quality assessment that can be used, for example, to select and fine-tune concatenation-based synthesizer element sets.>