On a method of preprocessing of sound base

摘要

In this report we describe the algorithm of sound fragments preprocessing, making sound base used in system of synthesis of speech under the text. Concatenative synthesis qualities are determined by quality of sound base in many respects. Each period has some initial phase, and periods, following one by one, not necessarily have identical initial phases by numbering a speech signal. Presence of initial phases creates problems at the analysis and updating, both the separate periods, and sounds as a whole. It is required to result a signal in a zero phase on frequency of the basic tone in the beginning of each period for elimination of distortions on joints of the periods and reductions of the distortions arising at their updating. Moreover values of last sample in all periods should coincide. Described preprocessing of sound base allows reduce considerably the high-frequency distortions arising at updating of sound fragments. The sound base containing speech fragments, given to zero initial phases, allows receive synthesized speech with higher quality signal, than the sound base containing speech fragments with casual initial phases.