本文選題：音頻特征 + 高斯混合模型��； 參考：《電子科技大學(xué)》2014年碩士論文

【摘要】：要大力實現(xiàn)交通運輸系統(tǒng)的智能化,智能交通運輸系統(tǒng)的發(fā)展至關(guān)重要,其關(guān)鍵在于車輛的檢測及識別。而目前所使用的主流檢測方法,由于種種原因,尚難以滿足沿道路大量設(shè)置的要求,因此,本文以車輛靜止或行駛時產(chǎn)生的音頻信號為基礎(chǔ),主要對車輛音頻信號的特征進行分析,在此基礎(chǔ)上提出基于車輛音頻信號對車型進行識別的方案的理論研究,并進行初步識別。(1)闡述軟件開發(fā)的系統(tǒng)總體設(shè)計方案,詳細描述系統(tǒng)整體的軟件架構(gòu),主要介紹系統(tǒng)設(shè)計中涉及到的音頻特征提取模塊和基于音頻特征的車輛識別模塊設(shè)計思路,并介紹系統(tǒng)的數(shù)據(jù)庫構(gòu)成。(2)主要完成音頻特征提取模塊的詳細設(shè)計。首先探討音頻去噪方法,為了提高算法對不同信噪比的帶噪音頻的處理能力,結(jié)合維納濾波和自適應(yīng)濾波的優(yōu)勢,對譜減法進行改進;其次針對本文所設(shè)計的系統(tǒng)平臺,為保證系統(tǒng)識別性能,充分考慮音頻幀內(nèi)和幀間的信息,探討選擇Mel倒譜與一階差分Mel倒譜作為特征參數(shù),在一定程度上提高系統(tǒng)的穩(wěn)健性。(3)主要探討識別模型的設(shè)計方面,深入研究概率模型中的高斯混合模型方法,GMM不僅能利用到音頻信號的時序動態(tài)信息,端點檢測的精度對其識別性能的影響也很小,設(shè)計實現(xiàn)基于GMM的車型識別方法。(4)對前述各章中設(shè)計的軟件功能和檢索算法分別進行了試驗測試和軟件測試,進行了仿真實驗,并得出了相關(guān)結(jié)論。
[Abstract]:In order to realize the intelligence of transportation system, the development of intelligent transportation system is very important, and the key lies in the detection and identification of vehicles. However, the mainstream detection methods used at present are still difficult to meet the requirements of a large number of settings along the road due to various reasons. Therefore, this paper is based on the audio signals produced by vehicles at rest or driving. Based on the analysis of the characteristics of the vehicle audio signal, this paper puts forward the theoretical research of the vehicle model recognition based on the vehicle audio signal, and describes the overall system design scheme of the software development. The software architecture of the whole system is described in detail. The design ideas of audio feature extraction module and vehicle recognition module based on audio feature are introduced. This paper introduces the database structure of the system. It mainly designs the audio feature extraction module. In order to improve the processing ability of the algorithm with different SNR, combining the advantages of Wiener filter and adaptive filter, the spectral subtraction method is improved. In order to ensure the recognition performance of the system, the information in and between the audio frames is fully considered, and the Mel cepstrum and the first-order differential Mel cepstrum are discussed as the feature parameters. In order to improve the robustness of the system to some extent, the design of the recognition model is mainly discussed. The Gao Si hybrid model method in probabilistic model can not only make use of the timing dynamic information of audio signal, but also have little effect on the performance of endpoint detection. The design and implementation of the vehicle recognition method based on GMM. 4) the software function and retrieval algorithm designed in the above chapters are tested and tested respectively, and the simulation experiments are carried out, and the relevant conclusions are obtained.
【學(xué)位授予單位】：電子科技大學(xué)
【學(xué)位級別】：碩士
【學(xué)位授予年份】：2014
【分類號】：TN912.34;TP311.52

【參考文獻】

相關(guān)期刊論文 前3條

1 朱學(xué)芳,徐建平;計算機語音信號處理與語音識別系統(tǒng)[J];南京郵電學(xué)院學(xué)報;1998年Z1期

2 韋曉東,胡光銳,任曉林;應(yīng)用倒譜特征的帶噪語音端點檢測方法[J];上海交通大學(xué)學(xué)報;2000年02期

3 王晶,傅豐林,張運偉;語音增強算法綜述[J];聲學(xué)與電子工程;2005年01期

相關(guān)碩士學(xué)位論文 前1條

1 陳國明;基于人耳掩蔽效應(yīng)的語音增強算法研究[D];東南大學(xué);2005年

，

本文編號：1969330