汉语语音识别系统中关键词检测技术的研究
发布时间：2019-01-18 08:43
【摘要】：语音识别技术是实现计算机与人之间进行语言交流的关键技术之一,语音识别技术在交互方面具有极高的应用价值。随着关键词检测、语音合成、声纹识别等技术的进步,人们对语音技术的需求逐渐加深,语音识别产品正在走进我们的工作生活,逐渐深入到各个领域。 本文主要是对汉语语音识别系统中关键词技术进行研究,通过两类不同的实际应用场景对关键词检测技术进行验证,主要的工作成果如下： 1、针对智能电话客服的应用深入研究了基于垃圾模型(Garbage Model)关键词检测技术。完成了端点检测技术的改进、垃圾语料建模选取、以及确定垃圾模型和关键词模型网络权重值,再此基础上与人合作实现了关键词检测引擎,通过对接国内某电脑生产商维修预约客服电话系统,实现对关键词检测技术的验证,该系统中典型的关键词有硬盘问题、开机故障、系统崩溃等。通过话术引导、合理增加关键词长度或者合并关键词显著提高关键词检出率。 2、面向语音文档检索应用需求,系统研究了基于音节网络(Syllable)的关键词检测技术,完成了对保险业和旅游业的的声学模型的训练,引入语言学知识提高关键词识别效果。借助保险业和旅游业的客服录音记录进行测试表明：可依据预先设定的关键词实现语音文档检索功能。
[Abstract]:Speech recognition technology is one of the key technologies to realize the language communication between computers and people. Speech recognition technology has a high application value in the aspect of interaction. With the development of key word detection, speech synthesis, voiceprint recognition and other technologies, the demand for speech technology is deepening. Speech recognition products are coming into our work and life, and gradually into every field. This paper mainly studies the keyword technology in the Chinese speech recognition system, and verifies the keyword detection technology through two different kinds of practical application scenes. The main work results are as follows: 1. Aiming at the application of smart phone customer service, the (Garbage Model) keyword detection technology based on garbage model is studied. The improvement of endpoint detection technology, the selection of garbage corpus modeling, and the determination of network weights of garbage model and keyword model are completed. On this basis, the keyword detection engine is implemented in cooperation with people. By docking a domestic computer manufacturer's maintenance reservation customer service telephone system, the key words in the system can be verified. The typical keywords in the system are hard disk problem, boot failure, system collapse and so on. Through the guidance of words, the reasonable increase of keyword length or the combination of keywords significantly improve the detection rate of keywords. 2. In order to meet the requirement of speech document retrieval, the keyword detection technology based on syllable network (Syllable) is studied systematically. The acoustic model of insurance industry and tourism industry is trained, and linguistic knowledge is introduced to improve the effect of keyword recognition. With the help of customer service recording of insurance industry and tourism industry, it is shown that voice document retrieval can be realized according to predefined keywords.
【学位授予单位】：北京邮电大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TN912.34
本文编号：2410518
[Abstract]:Speech recognition technology is one of the key technologies to realize the language communication between computers and people. Speech recognition technology has a high application value in the aspect of interaction. With the development of key word detection, speech synthesis, voiceprint recognition and other technologies, the demand for speech technology is deepening. Speech recognition products are coming into our work and life, and gradually into every field. This paper mainly studies the keyword technology in the Chinese speech recognition system, and verifies the keyword detection technology through two different kinds of practical application scenes. The main work results are as follows: 1. Aiming at the application of smart phone customer service, the (Garbage Model) keyword detection technology based on garbage model is studied. The improvement of endpoint detection technology, the selection of garbage corpus modeling, and the determination of network weights of garbage model and keyword model are completed. On this basis, the keyword detection engine is implemented in cooperation with people. By docking a domestic computer manufacturer's maintenance reservation customer service telephone system, the key words in the system can be verified. The typical keywords in the system are hard disk problem, boot failure, system collapse and so on. Through the guidance of words, the reasonable increase of keyword length or the combination of keywords significantly improve the detection rate of keywords. 2. In order to meet the requirement of speech document retrieval, the keyword detection technology based on syllable network (Syllable) is studied systematically. The acoustic model of insurance industry and tourism industry is trained, and linguistic knowledge is introduced to improve the effect of keyword recognition. With the help of customer service recording of insurance industry and tourism industry, it is shown that voice document retrieval can be realized according to predefined keywords.
【学位授予单位】：北京邮电大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TN912.34
【参考文献】
相关期刊论文 前10条
1 甄斌,吴玺宏,刘志敏,迟惠生;语音识别和说话人识别中各倒谱分量的相对重要性[J];北京大学学报(自然科学版);2001年03期
2 刘加;汉语大词汇量连续语音识别系统研究进展[J];电子学报;2000年01期
3 张鹏远;韩疆;颜永红;;关键词检测系统中基于音素网格的置信度计算[J];电子与信息学报;2007年09期
4 马晓梅;沈洁;;垃圾模型技术在关键词检测系统中的应用[J];信息技术;2009年06期
5 荣薇;陶智;顾济华;赵鹤鸣;;基于改进LPCC和MFCC的汉语耳语音识别[J];计算机工程与应用;2007年30期
6 刘鑫,陆林生;关键词检测系统中声学置信度的应用[J];计算机工程;2004年08期
7 幸璐璐,谢莎莎,孙甲松,王作英;在线废料建模在特定领域语音识别中的应用[J];计算机应用;2005年12期
8 倪崇嘉;刘文举;徐波;;汉语大词汇量连续语音识别系统研究进展[J];中文信息学报;2009年01期
9 宋战江,郑方,徐明星,武健,吴文虎;汉语连续语音识别系统与知识导引的搜索策略研究[J];自动化学报;2000年04期
10 孟猛;王晓瑞;梁家恩;徐波;;一种基于互补声学模型的多系统融合语音关键词检测方法[J];自动化学报;2009年01期
相关博士学位论文 前2条
1 李伟;基于内容的汉语语音检索技术研究与系统实现[D];清华大学;2011年
2 李宝祥;语音关键词检索若干问题的研究[D];北京邮电大学;2013年
，本文编号：2410518
本文链接：https://www.wllwen.com/kejilunwen/wltx/2410518.html
