Create An Emotional Speech Synthesis Corpus

Emotional speech synthesis as a new direction of a speech synthesis, which isinvolved in physiology, psychology, linguistics, information science and otherdisciplines of knowledge, can be applied to text reading, publish information querysystem, computer aided instruction and so on，the organic integration of speakingconvergence of voice analysis, sentiment analysis and computer technology to realizepeople-oriented and establish the foundation for the speech synthesis system withpersonalized features. Currently method of emotional speech synthesis includesrule-based synthesis and splicing-based waveform synthesis. The paper analyzes theacoustic characteristics of speech emotion， puts forward the emotional speechsynthesis model to realize the change of input neutral voice to the targetemotion sound.1. The dissertation introduced the3-D PAD emotion model to represent emotionscontained in speech quantitatively.2. The dissertation proposed an emotional speech corpus building method,which base on "the most natural control" theory and put forward by scene video torecord emotional data. For ensuring the effectiveness of the quality of recording data,the paper took three measures: designed high quality texts in accordance withthe emotional expression; analyzed emotion through the changes ofphysiological signal; carefully designed film material to stimulate the emotion fromauditory and visual aspects.3. The introduction of PRAAT analysis tool, which analyze and extract sentences,words and syllables layer3layers12dimensional features, create database ofcharacteristic parameters.4. Comparing the advantages and disadvantages of BP and RBF neuralnetwork in the training parameters corpus, the paper present the integrated acoustic model and realized from input neutral voice to the target emotion sound by analyzingstatistical parametric and unit selection speech synthesis.Study on emotional speech synthesis is divided into emotional analysis andspeech synthesis of two parts. The main work of emotional analysis, collect speechdata from different emotional acoustic, features extraction, analysis of acousticcharacteristics and emotional connection; the main work of speech synthesis is toestablish emotional conversion model, using the model for synthesis.