Emotional Voice Conversion Based On Pitch Target Model And Modified Prosody Parameters

The human voice not only contains the linguistic information, it also contains the extra-linguistic information. The linguistic information aims to express the text. The extra-linguistic information, which has nothing to do with the language and speech, has a great relationship with the speaker’s attitude and emotion. The human’s voice carries a wealth of emotional contents. The same text with different tones, different accents, can express different emotions and give others different feelings. Thus, emotion is an important part of the voice. There is a very important practical significance of research on the emotion speech.The emotional voice conversion is the conversion of the source voice signal’s prosody parameters and spectrum parameters to the target voice emotional prosody parameters and spectrum parameters, and synthesize the voice of the target emotion.In order to synthesize the target emotional states of the voice, we propose a conversion method based on Pitch Target model and modified prosody parameters. For happy, angry, sad and neutral four different kinds of emotions, we establish the syllable-based parameter library of the Pitch Target model. We model the speech emotional prosody parameters, and get the functions of the conversion of the neutral speech to the emotional speech. Then we analysis the emotional speech prosody characteristics parameter in the speech database, and change the converted emotional speech prosody parameter. In the end, we use the STRAIGHT model to synthesize the speech with emotional color.