基于可穿戴社交感知系统的语音分割算法研究

发布时间：2018-06-18 03:20

本文选题：HMM算法 + HMM-KLD算法　；参考：《电子科技大学》2017年硕士论文

【摘要】：随着科技的进步和人们生活水平的提高,身心健康成为当今社会的关注问题。通常,研究者可以通过社交感知特征客观分析和评估身心健康状态。语音信号处理是该领域重要的研究方向。它可以通过提取分析和融合语音特征客观地综合评估社交人群的心理健康。因此,高效的语音分割算法将有利于社交语音感知特征的提取。本文将针对传统非监督语音分割算法中基于HMM(Hidden Markov Model)分割精度的局限,提出了融合KL散度(kullback-leibler divergence)的HMM语音分割算法,进一步提高语音分割准确性。但是,从HMM-KLD的语音分割结果中,存在分割算法最优判定问题。对于该难点,本文提出了基于稀疏性相关特征的自动判别语音分割方法,有效地解决这一难题。本文将从以下几个方面进行具体阐述。(1)基于可穿戴社交感知系统和语音分割系统的国内外研究现状,本文提出了了高效的语音分割算法可以较好地帮助社交感知系统进行语音特征分析并有利于研究在实际应用中社交感知行为或心理状态与语音特征的关联。(2)基于传统的HMM非监督语音分割方法,评估在不同噪音场景下的语音分割准确性。这一过程主要分为三个阶段,第一阶段是基于谱熵和短时自相关特征进行去噪;第二阶段基于短时能量去除无声音的信号部分;第三阶段主要是基于不同穿戴社交感知设备的人的短时能量的不同,分割出穿戴者语音信号。(3)由于HMM算法分割精度的局限性,本文提出了融合KL散(kullback-leibler divergence)的HMM语音分割算法可以进一步提高语音分割准确性,并通过所采集的语音信号验证新算法的改进效果。(4)根据HMM-KLD的语音分割算法中存下的最优判定问题,基于语音稀疏性相关特征,本文提出了一种可以自动优化判别语音分割算法的策略,进一步提高语音分割算法的准确性。(5)基于基本语音特征和韵律语音特征,本文探索分析说话人之间亲密度以及老年群体的社交特征与他们的语音感知特征的联系。(6)总结全文和展望未来,主要总结本文中的语音分割算法的优劣性和后期可改进的一些方案,同时展望将来语音分割算法以及语音特征分析在可穿戴社交感知系统中的应用。
[Abstract]:With the progress of science and technology and the improvement of people's living standards, physical and mental health has become a social concern. Usually, researchers can objectively analyze and evaluate the state of physical and mental health through the characteristics of social perception. Speech signal processing is an important research direction in this field. It can objectively and synthetically evaluate the mental health of social groups by extracting and analyzing and integrating speech features. Therefore, efficient speech segmentation algorithm will be conducive to feature extraction of social speech perception. In this paper, aiming at the limitation of traditional unsupervised speech segmentation algorithm based on hmm Hidden Markov Model, a hmm speech segmentation algorithm based on KL divergence and Kullback-leibler divergence is proposed to further improve the accuracy of speech segmentation. However, from the HMM-KLD speech segmentation results, there is an optimal decision problem of segmentation algorithm. For this difficulty, an automatic discriminant speech segmentation method based on sparse correlation features is proposed to solve this problem effectively. In this paper, the following aspects of the specific elaboration of the following aspects of the wearable social perception system and speech segmentation system based on the domestic and foreign research status, In this paper, an efficient speech segmentation algorithm is proposed, which can be used to analyze speech features in social perception systems and to study the relationship between social perception behavior or mental state and speech features in practical applications. Traditional hmm unsupervised speech segmentation method, To evaluate the accuracy of speech segmentation in different noise scenes. This process is mainly divided into three stages: the first stage is based on spectral entropy and short-time autocorrelation, the second stage is based on short-term energy to remove the soundless signal. The third stage is mainly based on the difference of short-term energy of people wearing different social perception devices, and the speech signal of the wearer is segmented. (3) because of the limitation of the segmentation accuracy of hmm algorithm, In this paper, we propose a speech segmentation algorithm based on KL scattered Kullback-leibler divergence (hmm), which can further improve the accuracy of speech segmentation. The improved effect of the new algorithm is verified by the collected speech signals. (4) according to the optimal decision problem in HMM-KLD 's speech segmentation algorithm, Based on the features of speech sparsity, this paper proposes a strategy to automatically optimize speech segmentation algorithm, which can further improve the accuracy of speech segmentation algorithm. It is based on basic speech features and prosodic speech features. This paper explores the relationship between the speaker's affinity and the social characteristics of the elderly and their phonological perception. (6) summing up the full text and looking forward to the future. This paper mainly summarizes the advantages and disadvantages of the speech segmentation algorithm in this paper and some schemes that can be improved in the later stage. At the same time, it looks forward to the application of speech segmentation algorithm and speech feature analysis in wearable social perception system in the future.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN912.3

【参考文献】