东北大学本科毕业设计（论文） Abstract
Design and Implementation of Individuation Text-Voice
Conversion (TTS) System
Abstract
The voice technology is a new science and technology nearly half a century,
mainly including speech recognition and speech synthesis technology. Among them,
speech synthesis converts text into sound signals that human can understand. In fact,
text-voice conversion (abbreviated TTS) has very broad applications and it can
convert any text to voice signals. Thus, it has broad application prospects in car
navigation, telecommunications call and the bus train station names automatically
broadcast.
However, current TTS system only can speak text in system roles without any
emotion, so there is a little lack in personalization elements for user experience.
Therfore, after voice feature analysis, this article sets up personalization voice
convertion functions. So users can set any speaking roles by themselves. It also
establishes theoretical foundation for speech signal personalized processing.
Firstly, this article expounds TTS basic functions focused on Microsoft Speech
SDK. Then, using the API interface of Speech SDK, we can let the computer read the
text and achieve mixed reading in English and Chinese. Next, to get personalization,
we use a voice conversion model. Voice Conversion is for the source speaker and the
target speaker, namely converting a source speaker speech into a speech with target
speaker acoustic features. In addition, this article talks about voice emotion factors.
Extracting and analyzing different emotional speech feature parameters to get
conversion relationship among different emotions and neutral speech. Finally, we can
get a speech with emotional features meeting user need.
Finally, this article designs a personalization text-voice conversion system with
C++ and Matlab mixed programming. It reads text in both system roles and customed
roles. The speaking result is acceptable and the effectiveness of the voice conversion
algorithm is demonstrated.
评论0