【摘要】：近年來(lái)隨著對(duì)情感計(jì)算不斷地研究,語(yǔ)音情感識(shí)別得到了研究者們廣泛的關(guān)注,它的實(shí)現(xiàn)對(duì)于推動(dòng)心理學(xué)發(fā)展,構(gòu)建更加和諧的人機(jī)環(huán)境起到非常重要的作用。語(yǔ)音情感識(shí)別是指通過(guò)提取語(yǔ)音中與情感相關(guān)聯(lián)的特征參數(shù),將這些特征參數(shù)組成特征向量,使用分類模型對(duì)特征向量進(jìn)行計(jì)算,最終分析出情感類別。其中不斷提高分類模型的識(shí)別性能一直是研究者們研究的重點(diǎn)。 為了提高識(shí)別性能,本文提出了基于深度信念網(wǎng)絡(luò)的語(yǔ)音情感識(shí)別策略,深度信念網(wǎng)絡(luò)通過(guò)構(gòu)建多隱層的人工神經(jīng)網(wǎng)絡(luò),以此達(dá)到高效的特征學(xué)習(xí)能力,彌補(bǔ)了傳統(tǒng)的神經(jīng)網(wǎng)絡(luò)在特征選擇方面以及對(duì)于復(fù)雜函數(shù)的表示能力有限的缺點(diǎn),提高了對(duì)于復(fù)雜分類問(wèn)題的泛化能力,同時(shí)也降低了神經(jīng)網(wǎng)絡(luò)訓(xùn)練的收斂時(shí)間,最終使識(shí)別性能得到了提高。本文使用MATLAB實(shí)現(xiàn)了基于深度信念網(wǎng)絡(luò)的語(yǔ)音情感識(shí)別策略,通過(guò)收集語(yǔ)音情感數(shù)據(jù)集,將該策略同基于BP神經(jīng)網(wǎng)絡(luò)分類模型的語(yǔ)音情感識(shí)別方法進(jìn)行對(duì)比,分析召回率,準(zhǔn)確率以及F1值三個(gè)指標(biāo)。通過(guò)一系列實(shí)驗(yàn)顯示,本文所提出的策略在平均召回率、平均準(zhǔn)確率以及F1值均比BP神經(jīng)網(wǎng)絡(luò)要高。 基于本策略,本文開(kāi)發(fā)了一款移動(dòng)語(yǔ)音情感識(shí)別系統(tǒng)原型,該系統(tǒng)原型采用C/S架構(gòu),客戶端主要有錄音、語(yǔ)音播放、上傳語(yǔ)音以及結(jié)果顯示等功能,服務(wù)器端主要有特征參數(shù)提取以及情感識(shí)別等功能。用戶通過(guò)麥克風(fēng)錄取自己的語(yǔ)音,然后上傳到服務(wù)器進(jìn)行語(yǔ)音分析,服務(wù)器最終將情感識(shí)別結(jié)果返回給客戶端。
[Abstract]:In recent years, with the continuous research on emotional computing, speech emotion recognition has been widely concerned by researchers. Its realization plays a very important role in promoting the development of psychology and building a more harmonious human-computer environment. Speech emotion recognition means that by extracting the feature parameters associated with emotion in speech, these feature parameters are formed into feature vectors, and then the feature vectors are calculated by classification model, and finally the emotion categories are analyzed. Among them, improving the recognition performance of classification models has been the focus of researchers. In order to improve the recognition performance, this paper proposes a speech emotion recognition strategy based on the deep belief network. The deep belief network constructs a multi-hidden layer artificial neural network to achieve an efficient feature learning ability. It makes up for the shortcomings of the traditional neural network in feature selection and the limited representation of complex functions, improves the generalization ability for complex classification problems, and reduces the convergence time of neural network training. Finally, the recognition performance is improved. This paper uses MATLAB to realize speech emotion recognition strategy based on deep belief network. By collecting speech emotion data set, the strategy is compared with speech emotion recognition method based on BP neural network classification model, and the recall rate is analyzed. The accuracy rate and F1 value are three indexes. A series of experiments show that the average recall rate, average accuracy rate and F1 value of the proposed strategy are higher than those of BP neural network. Based on this strategy, this paper develops a mobile speech emotion recognition system prototype, which uses C / S architecture. The client has the functions of recording, voice playing, uploading voice and displaying results. The server has the function of feature parameter extraction and emotion recognition. The user records his voice through microphone, then uploads it to the server for voice analysis, and the server finally returns the result of emotion recognition to the client.
【學(xué)位授予單位】：大連理工大學(xué)
【學(xué)位級(jí)別】：碩士
【學(xué)位授予年份】：2014
【分類號(hào)】：TN912.3

【參考文獻(xiàn)】

相關(guān)期刊論文 前2條

1 余伶俐;蔡自興;陳明義;;語(yǔ)音信號(hào)的情感特征分析與識(shí)別研究綜述[J];電路與系統(tǒng)學(xué)報(bào);2007年04期

2 趙臘生;張強(qiáng);魏小鵬;;語(yǔ)音情感識(shí)別研究進(jìn)展[J];計(jì)算機(jī)應(yīng)用研究;2009年02期

，

本文編號(hào)：2335952