论文部分内容阅读
对于合成语音的音质进行评价，是评价合成系统使用价值和性能的一种主要方式．通过音质评价，找出影响音质的原因，反过来指导音质改善，这正好体现了人类发声时听觉的反馈作用．现行的评价方案多为主观评价，不仅不利操作，而且不能对改善音质提供定量的建议．为了实现对语音音质的客观评价，作者通过对LSP参数的频谱敏感度的研究，将其与听觉特性相联系，用频谱敏感度表征语音中对听觉敏感的成分，从而给出了符合人耳听觉特性的音质的定量描述，提供了一种对语音音质的客观评价方法．并通过实验验证了这种方法的正确性．
Evaluating the sound quality of synthesized speech is a major way to evaluate the value and performance of a synthesis system. Through the evaluation of the sound quality, find out the reasons that affect the sound quality, in turn, to guide the sound quality to improve, which just shows the human auditory feedback function. The current evaluation program mostly subjective evaluation, not only unfavorable operation, and can not provide quantitative advice to improve the sound quality. In order to achieve an objective evaluation of the voice quality, the author, by studying the spectral sensitivity of the LSP parameters, associates it with the auditory properties and characterizes the auditory-sensitive components of the speech using spectral sensitivity, The quantitative description of the characteristic sound quality provides a objective assessment of the sound quality of the speech. The correctness of this method is verified by experiments.