基于DTW的语音关键词检出
发布时间：2018-07-24 19:03
【摘要】：针对少资源语言的语音关键词检出技术受到了广泛关注。该文在基于动态时间规整(dynamic time warping,DTW)的关键词检出框架下,提出了基于音素边界的局部匹配策略,用以解决基于样例的语音关键词检出任务中的近似查询问题。在QUESST 2014评测数据上采用多种特征进行了实验验证。实验结果显示:基于音素边界的局部匹配策略不仅在近似查询T2和T3任务上的检出效果明显提升,在精确查询T1任务上也获得了有效提升。随后的系统融合实验表明,该策略能够大幅提升融合系统的性能。
[Abstract]:Speech keyword detection technology for less resource languages has been paid more and more attention. In this paper, a local matching strategy based on phoneme boundary is proposed under the framework of keyword detection based on dynamic time warping (dynamic time (dynamic time warping (dynamic time), which is used to solve the approximate query problem in speech keyword detection task based on sample. A variety of features are used to verify the QUESST 2014 data. The experimental results show that the local matching strategy based on phoneme boundary not only improves the detection effect on approximate T _ 2 and T _ 3 tasks, but also improves the precision query T _ 1 tasks. Subsequent system fusion experiments show that the strategy can greatly improve the performance of the fusion system.
【作者单位】： 西北工业大学计算机学院 陕西省语音与图像信息处理重点实验室;南洋理工大学Temasek实验室;新加坡科技局资讯通信研究院 人类语言技术部;南洋理工大学计算机工程学院;
【基金】：国家自然科学基金面上项目(61571363)
【分类号】：TP391.3
本文编号：2142341
[Abstract]:Speech keyword detection technology for less resource languages has been paid more and more attention. In this paper, a local matching strategy based on phoneme boundary is proposed under the framework of keyword detection based on dynamic time warping (dynamic time (dynamic time warping (dynamic time), which is used to solve the approximate query problem in speech keyword detection task based on sample. A variety of features are used to verify the QUESST 2014 data. The experimental results show that the local matching strategy based on phoneme boundary not only improves the detection effect on approximate T _ 2 and T _ 3 tasks, but also improves the precision query T _ 1 tasks. Subsequent system fusion experiments show that the strategy can greatly improve the performance of the fusion system.
【作者单位】： 西北工业大学计算机学院 陕西省语音与图像信息处理重点实验室;南洋理工大学Temasek实验室;新加坡科技局资讯通信研究院 人类语言技术部;南洋理工大学计算机工程学院;
【基金】：国家自然科学基金面上项目(61571363)
【分类号】：TP391.3
【相似文献】
相关重要报纸文章 前2条
1 叶向荣 郑民军;大破语音关[N];中国教育报;2001年
2 山东省阳信县翟王镇韩箔中心小学 张玉兰;语文课堂 让学生快乐表达[N];学知报;2010年
相关硕士学位论文 前1条
1 朱林;语音关键词检测若干问题研究[D];北京邮电大学;2015年
，本文编号：2142341
本文链接：https://www.wllwen.com/kejilunwen/ruanjiangongchenglunwen/2142341.html
