论文部分内容阅读
国际上对自动语种识别进行了广泛的研究,提出了各种各样的方法,美国国家标准技术研究所(NIST)多年的评测表明,基于并行音素识别(parallel phoneme recognition language modeling,PPRLM)的方法取得了很好的性能。该文提出了一种基于多种语言的音素识别方法的自动语种识别系统,系统中Multilingual音素集是使用基于数据驱动聚类获得。通过真实环境电话语音测试表明,该方法在只使用了很少的识别时间的情况下,获得了跟传统的PPRLM系统可比的识别正确率。同时经过与PPRLM系统融合后,获得了更好的性能,跟其他主流的几种语种识别方法也有可比的性能。
Internationally, automatic language recognition has been extensively studied and a variety of methods have been proposed. The years of evaluation conducted by the National Institute of Standards and Technology (NIST) show that the method based on parallel phoneme recognition language modeling (PPRLM) Achieved good performance. In this paper, an automatic speech recognition system based on multi-lingual phoneme recognition is proposed. Multilingual phoneme sets in the system are obtained using data-driven clustering. Real-world telephone voice tests show that this method achieves comparable recognition accuracy with the traditional PPRLM system with only a small amount of recognition time. At the same time, better performance has been achieved after integration with the PPRLM system, with comparable performance to several other mainstream language recognition methods.