|其他题名||research on technology and application of human-machine speech dialogue systems
Human-machine Speech Dialogue Technology
|中文摘要||人机语音对话系统是指在用户说话方式，说话内容受限较小的情况下，系统能够准确的识别用户提到的关键词语，理解用户的意图，并做出合理的应答，使整个人机对话过程自然流畅的进行。 根据人机语音对话系统的特点，本文以语音识别技术为基础，重点研究了搜索算法、对话管理和端点检测技术，以提高人机对话系统的识别率和灵活性。同时完善和扩展了大规模宽领域语料库的对话规则，添加和整理了大量的限定领域语料和语音数据，开发了基于整词建模的特定人和非特定人的孤立词识别系统。 本文开展的工作主要集中于几个不同的人机语音对话系统的技术和应用研究。论文的主要工作概括如下： 1. 改进了基于关键词检测技术的人机语音对话系统。结合系统特点，改进了搜索算法，搜集整理了限定领域语料，对语料库中的大规模宽领域语料进行完善和扩展，添加和改进了对话管理规则，使人机对话更加流畅。 2. 开发了基于整词建模的特定人和非特定人的孤立词识别系统。特定人系统可以方便的更改命令词表、在线训练、在线识别，方便用户使用。同时，采集整理了大量的语音数据，并对两个系统做了大量的对比实验。 3. 改进了语音识别应用中的端点检测方法。根据对自适应技术的研究，提出了一种改进的基于子带自适应谱熵的端点检测方法，该方法可以很好的区分语音和噪音，增加系统的鲁棒性，提高系统识别率。 4. 实现了基于嵌入式DSP模块的几个应用系统。该模块用的是基于命令词的语音识别技术，并应用到语音玩具，智能家居以及移动机器人中。|
|英文摘要||Human-machine speech dialogue system can recognize users’ speech, understand users’ intent and make a reasonable response with less limited tongue, and then thus a natural and fluent dialog process can be achieved. According to the characteristics of the human-machine speech dialogue system, this paper, based on speech recognition technology, focuses on the research of search algorithm, dialogue management and endpoint detection technology to improve the recognition rate and flexibility of human-machine dialogue system. Additionally, rules of the large-scale, wide-ranging field corpus have been improved and expanded, a lot of corpus and voice data of limited-field has been added and organized, and a speaker-dependent and speaker-independent isolated words recognition system have been developed based on whole-word modeling. In this paper we focus on the research on techniques and application of several different human-machine dialogue systems. The main tasks of this paper include the following: 1. Improve the human-machine speech dialogue system which is based on the key words detection technology. According to the characteristic of this system, the search algorithm has been improved, and additionally, corpus of the limited-field have been collected and classified, corpus of the large-scale, wide-ranging field have been improved and expanded, and dialogue management rules have been added to achieve a fluent human-machine dialogue process. 2. Develop a speaker-dependent and speaker-independent isolated words recognition system both based on whole-word modeling. The speaker-dependent system can change command word lists, train model and recognize online conveniently and user-friendly. Additionally, amount of voice data has been collected and classified, and a lot of comparative experiments have been made on both two systems. 3. Improve the endpoint detection algorithm in application of speech recognition technology. Based on the research of adaptive technology, an improved speech endpoint detection algorithm based on adaptive band-partition spectral entropy is proposed. This algorithm is advantageous in discriminating noise from the speech voice, and increases the robust of system and recognition rate. 4. Implement several practical application systems based on embedded DSP module which is based on command words detection technology and has been applied in voice toys, smart home system and mobile robots.|
王琳. 人机语音对话系统的技术与应用研究[D]. 中国科学院自动化研究所. 中国科学院研究生院,2010.