基于HMM的关键词语音识别技术在智能家居中的应用研究

发布时间：2018-06-28 23:16

本文选题：智能家居 + 语音识别　；参考：《兰州交通大学》2017年硕士论文

【摘要】：智能家居中使用语音识别控制技术具有方便快捷的特点,可以摆脱手动遥控带来的繁琐并解决智能化体验差的问题,满足人民大众对家居生活智能化越来越高的需求。实际应用中家居系统对语音识别算法的识别率和识别效率要求较高,由于基于神经网络算法的语音识别技术需依托因特网资源以及网络高性能服务器平台,无法达到智能家居对硬件低配,算法轻量的要求。目前大部分智能家居设备中的语音控制系统主要采用基于动态规整算法(DTW)的孤立词识别系统、基于隐马尔科夫模型(HMM)的孤立词识别系统以及部分高配置硬件系统采用的基于隐马尔科夫填料模板的关键词识别系统,这些算法存在识别率偏低或算法计算量大、效率偏低、不适于低配硬件环境以及孤立词识别技术体验效果差的问题。因此对智能家居系统中语音识别技术进行算法改进或者方案改进,保证同时满足较高的算法识别速率和识别率,以满足用户对智能家居语音控制的体验效果,对进一步开发智能家居系统具有重要的意义。首先对语音识别技术的基本原理进行了详细介绍和分析,包括预处理、特征提取、模板训练和模式匹配四个环节。其次详细介绍了目前智能家居中常用的轻量级语音识别算法,对算法优点以及不足对比分析。对DTW和HMM分别对应的孤立词识别系统和关键词识别系统详细介绍和对比分析,根据不足给出改进的关键词识别算法。再次利用MATLAB工具进行系统仿真,测试算法可行性、识别率、识别速率等指标。最后利用B/S模式来设计开发一套基于Web平台的关键词语音识别系统。本文主要解决方案是在预处理阶段,在传统的低通滤波去噪基础上,加入小波去噪进行语音增强并抑制非平稳噪声。在特征提取以及端点检测阶段将传统的端点检测算法改为单音节端点检测分割算法。识别阶段滑动匹配所有音节,将得分最高的提取出来,达到确认关键词得分阈值即为关键词。对改进的算法进行MATLAB仿真,关键词识别率和识别速率基本达到系统要求,体验效果得到改善,在一定程度上提高了智能家居语音控制智能化程度。对于系统存在的问题,提出了未来可以继续改进的方向。
[Abstract]:The use of speech recognition control technology in smart home has the characteristics of convenience and shortcut. It can get rid of the complicated problem brought by manual remote control and solve the problem of intelligent experience, so as to meet the people's increasing demand for intelligent home life. In practical application, the recognition rate and recognition efficiency of the speech recognition algorithm are required by the home system in the actual application. Because the speech recognition technology based on neural network algorithm needs to rely on Internet resources and the high performance server platform of the network, it is unable to meet the requirements of low allocation of hardware and lightweight algorithm for smart home. At present, the speech control system in most intelligent home devices mainly adopts the isolated word recognition system based on dynamic normalization algorithm (DTW). Based on the hidden Markov model (HMM), a key word recognition system based on the hidden Markov stuffing template is adopted in the hidden Markov model (HMM). These algorithms have low recognition rate, large computation, low efficiency, unfit for low hardware environment and isolated word recognition technology. Therefore, the speech recognition technology in the smart home system is improved or improved to ensure that it meets the higher recognition rate and recognition rate at the same time so as to meet the user's experience effect on the intelligent home voice control. It is of great significance to the further development of intelligent home system. First, the speech recognition technology is used. The basic principle is introduced and analyzed in detail, including four links: preprocessing, feature extraction, template training and pattern matching. Secondly, the lightweight speech recognition algorithm commonly used in smart home is introduced in detail, and the advantages and shortcomings of the algorithm are analyzed. The isolated word recognition system and key word recognition corresponding to DTW and HMM classification are identified. The system is introduced and analyzed in detail, and the improved keyword recognition algorithm is given according to the shortage. MATLAB tool is used to simulate the system again, the feasibility of the algorithm, the recognition rate, the recognition rate and so on. Finally, a set of key word recognition system based on the Web platform is designed and developed by using the B/S model. The main solution is in this paper. In the preprocessing stage, on the basis of traditional low pass filter denoising, wavelet denoising is added to speech enhancement and suppression of non-stationary noise. In feature extraction and endpoint detection, the traditional endpoint detection algorithm is changed to monosyllabic endpoint detection segmentation algorithm. The recognition stage is sliding to match all syllables, and the highest score is extracted. To confirm the key word score threshold is the key word. MATLAB simulation of the improved algorithm, the key word recognition rate and recognition rate basically reach the system requirements, the experience effect is improved, to a certain extent, the intelligentized degree of the intelligent home voice control is improved. The future can be improved for the problems existing in the system. Direction.
【学位授予单位】：兰州交通大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN912.34;TU855

【参考文献】