Emotional Speech Conversion Using Spectrum-Prosody Dual Transformation

Natural speech not only includes the basic linguistics, but also carries the emotions.Speech with the same words may convey differernt informations if it carries differentemotions. Emotional speech conversion is to transform the emotin conveyed in the speechto target emotion while the keeping the same words and it has far-reaching significance.In this paper, we analyse the features of emotional speech based on the two publicemotional speech database: EMO-D and DES. Since spectral and prosodic features are keyfactors that influence the emotional effects of speech, this paper proposes aspectrum-prosody dual transformation method which is better than the traditional speechprocessing methods that only focuses on one and ignores another.Through analysing the merits and drawbacks of different spectrum features and themodels which can transform them, we choose the LSF as the spectrum features and chooseGMM (Gaussian Mixture Model) as the tansforming model in our spectrum transforationstage, and use the STRAIGHT to synthesize the emotional speech.In the prosody transformation stage, aiming at the time-varying character of prosodyfeatrues, we propose the PTR（Prosody Transformation Rule）to tansform them in each partof the speech which gain better effect than the traditional methods that make the wholesentence as the analysing unit. Since the stress can enhance the emotinal express of angryspeech and the fundamental frequency makes the greatest contribution, we propose thePTR combine the SGM (Single Gaussian Model) to transform the fundamental frequencyin the part o to make the angry speech transformed with stress to enhance the effect of theemotional conversion.At last, we evaluate the effect of the speech affter conversion from the subjective and objective respects, and the results show that the dual transformation method achieves agood effect with high subjective evaluating scores and78.25%objective Recognition rate.