Improved unit selection speech synthesis method utilizing subjective evaluation results on synthetic speech

Abstract

This paper presents an improved unit selection and waveform concatenation speech synthesis method by gathering and utilizing human feedbacks on synthetic speech. Firstly, a set of texts are synthesized by the baseline unit selection synthesis system. Each prosodic word within the synthetic speech is then evaluated as a natural one or an unnatural one by listeners. In our proposed method, these natural synthetic segments are treated as virtual candidate units to extend the original speech corpus for unit selection. A new speech synthesis system is constructed using this extended speech corpus. A synthetic error detector based on SVM classifier is also built using the natural and unnatural synthetic speech. At synthesis time, the input text is synthesized using the baseline system and the extended system simultaneously. The two unit selection results are evaluated by the trained synthetic error detector to determine the optimal one. Experimental results prove the effectiveness of our proposed method in improving the naturalness of synthetic speech on a task of synthesizing place names.