论文部分内容阅读
从1994年开始，对汉语语音合成系统的工作性能定期举行全国评测．采用语言清晰度测试方法，1994年对五个不同的合成系统进行了评测和诊断．听音人为16名大学生（男8，女8），对合成言语没有经验．听音人响应是开放的听音记录．同时，还采用十点主观评价（MOS）测定言语自然度．为给出各合成系统音段层的诊断信息，对合成语音的辅音知觉混淆矩阵进行了分析．借助于对比自然言语和合成言语在不同语言层次上清晰度试验得分间的统计关系，来考察合成系统韵律特征处理的缺陷．结果表明，采用上述方法可得到评测合成系统工作性能的稳定合理的指标．有关韵律特征的评价方法有待于进一步发展．
Since 1994, the national evaluation of the performance of Chinese speech synthesis system has been conducted regularly. Using a speech intelligibility test, five different synthetic systems were evaluated and diagnosed in 1994. Hearing students for 16 college students (male 8, female 8), no synthetic speech experience. Listener response is an open listening record. At the same time, ten subjective evaluations (MOS) are also used to measure the naturalness of speech. In order to give the diagnostic information of the sound layer of each synthetic system, the consonant perception confusion matrix of synthesized speech was analyzed. By comparing the statistical relationship between natural speech and synthetic speech in different linguistic levels of sharpness test scores, the defects of prosodic feature processing in synthetic systems are investigated. The results show that the above method can be used to evaluate the performance of a stable and reasonable synthetic system performance indicators. The evaluation method of prosodic features needs to be further developed.