Dimension reduction selection for increasing naturalness in speech synthesis

摘要

Speech synthesis systems are systems that provide great convenience to people on user-computer interaction. These systems can read written text as well as they can process human speech artificially. In this study instead of a monotone robotic speech synthesis, the idea is that it is better suited to human nature to have different types of texts sound differently in their own categories. For this process which is more suitable for human nature, the raw texts are first classified by passing through the text preprocessing step, The audio files are obtained by determining which audio texts are to be sounded with this tone as a result of this classification. Text speech that is not appropriate for its category is provided by differentiating its tonalities. Thus, misclassified documents can be clearly distinguished from tonal differences in audio files.