论文部分内容阅读
变异语音识别是一项极具挑战意义的研究课题,一种解决方法是在前端对语音进行变异分类,然后根据不同变异情况采用相关的处理算法。在各种语音变异中,说话人在战斗机、航天飞机座舱等环境中,身体受到应力（重力）影响时的情况更具有特殊性。其所引起的发音变异有别于因心理的、感知的或生理的因素所引起的变异,目前国内外还鲜见有关应力影响不变异语音分类问题的专门研究。木文从对应力影响下的几种基于基频的语音特征的分析出发,提出了对应力影响下的变异语音和正常语音进行分类的方法。对航空模拟飞行器中采集的小词表实验样本,特定人平均分类正确率达到了93.3％,多说话人分类上确率达到了85.8％。
Variational speech recognition is a challenging research topic. One solution is to classify the variations of speech at the front-end, and then use the relevant processing algorithms according to different variations. In various phonetic variations, the situation of the speaker under the influence of stress (gravity) in a fighter plane, space shuttle cockpit and other environments is even more peculiar. The variation of pronunciation caused by it is different from the variation caused by psychological, perceptual or physiological factors. At present, the special researches on the classification of stress-invariant variations in speech are rare. Based on the analysis of several fundamental frequency-based speech features under the influence of stress, Mu Wen proposed a method to classify the variant speech and the normal speech under the influence of stress. For the experimental sample of the small vocabulary collected in the aircraft, the average classification accuracy of the specific person reached 93.3% and the classification accuracy of the multi-speaker reached 85.8%.