A Unit Selection-based Speech Synthesis Approach for Chinese Mandarin Text-to-Speech

The paper presents a unit selection-based speech synthesis approach for Chinese Mandarin. Unit selection-based approach generates speech by directly connecting pre-recorded speech units. In this approach, a corpus is used as a source unit inventory. A feature vector is defined to describe each unit. To generate speech, the feature vector of the target unit is first calculated. During synthesis, we select units by considering the following: (1) each unit to be selected should have a proper prosody property (2) the adjacent units should be smoothly connected. To find the best unit sequence, Viterbi search algorithm is used in this approach. Prosody is one of the important concerns in speech synthesis. To generate natural speech, prosody should be carefully considered in unit selection process. In this paper, the authors use a prosody description to generate speech with good prosody. Experiment shows that this approach generates very high quality speech.