基于語音識別的家用服務(wù)機(jī)器人控制系統(tǒng)
發(fā)布時間：2018-08-28 08:15
【摘要】：本文主要研究語音識別的整個過程：主要由語音信號濾波、采樣、量化、加窗、端點檢測、特征提取、模型訓(xùn)練和閾值比較組成,以及通過Matlab實現(xiàn)對算法模型的仿真。同時通過Matlab的GUI設(shè)計技術(shù)實現(xiàn)了語音識別的交互界面。在語音識別理論的基礎(chǔ)上,通過搭建的五個自由度的Arduino雙臂機(jī)器人和ASR M08-A語音識別模塊,實現(xiàn)了語音控制機(jī)器人完成各種規(guī)劃動作。 語音信號經(jīng)過濾波、采樣與量化得到離散的數(shù)字信號后,進(jìn)行預(yù)加重,預(yù)加重的目的在于濾除低頻干擾,提升輸入信號的高頻分量。分幀使得原本的信號變成一段一段的,相當(dāng)于對原始信號時域內(nèi)加了一個矩形窗。時域內(nèi)與矩形窗相乘相當(dāng)于頻域內(nèi)信號頻譜與矩形窗的傅里葉變換進(jìn)行卷積。 雙門限端點檢測算法通過短時平均能量和過零率兩個門限來實現(xiàn)語音信號的端點檢測。實現(xiàn)語音信號的端點檢測后通過美爾頻率倒譜系數(shù)和一階差分美爾頻率倒譜系數(shù)獲得語音信號的特征參數(shù)。同時提出了改進(jìn)的語音信號特征參數(shù)提取算法,基于小波變換的線性預(yù)測倒譜系數(shù)的計算步驟,基于小波變換的美爾頻率倒譜系數(shù)的計算步驟。最后,利用基于小波變換的線性預(yù)測倒譜系數(shù)(DWTL)及相應(yīng)的差分參數(shù)(△DWTL)和基于小波變換的美爾頻率倒譜系數(shù)(DWTM)及相應(yīng)的差分參數(shù)(△DWTM)組成的系數(shù)矩陣。然后通過隱馬爾科夫模型當(dāng)中,前向后向算法、viterbi算法、Baum-welch算法實現(xiàn)模型訓(xùn)練。同時通過Matlab GUI設(shè)計以及回調(diào)函數(shù)的編寫實現(xiàn)了語音識別仿真交互界面。 在語音識別理論的基礎(chǔ)上,與Arduino雙臂機(jī)器人結(jié)合。五自由度服務(wù)機(jī)器人手臂通過坐標(biāo)通用旋轉(zhuǎn)變換算法實現(xiàn)機(jī)器人手臂的正運動學(xué)問題和逆運動學(xué)問題求解。正運動學(xué)問題是通過已知的機(jī)器人各個關(guān)節(jié)變量來求解末端執(zhí)行器的位姿；逆運動學(xué)問題根據(jù)機(jī)器人末端執(zhí)行器的位置和姿態(tài)要求,通過運動學(xué)逆解求得各個關(guān)節(jié)轉(zhuǎn)角。然后運用ASR M08-A語音識別模塊,32路舵機(jī)控制板、Arduino atmegal2560控制板、5自由度機(jī)械臂、實現(xiàn)語音控制機(jī)器人手臂動作。
[Abstract]:This paper mainly studies the whole process of speech recognition: it consists of speech signal filtering, sampling, quantization, windowing, endpoint detection, feature extraction, model training and threshold comparison, and the simulation of the algorithm model through Matlab. At the same time, the interactive interface of speech recognition is realized by GUI design technology of Matlab. On the basis of speech recognition theory, a five-degree-of-freedom Arduino dual-arm robot and a ASR M08-A speech recognition module are built to realize various planning actions of the speech control robot. After the speech signal is filtered, sampled and quantized, the discrete digital signal is pre-accentuated. The purpose of preemphasis is to filter out the low-frequency interference and enhance the high frequency component of the input signal. Framing makes the original signal a segment, which is equivalent to adding a rectangular window to the original signal. Multiplying with the rectangular window in the time domain is equivalent to convolution between the frequency spectrum of the signal and the Fourier transform of the rectangular window. The dual threshold endpoint detection algorithm realizes the endpoint detection of speech signal through two thresholds: the short time average energy and the zero crossing rate. After the endpoint detection of speech signal is realized, the characteristic parameters of speech signal are obtained by the number of Mel frequency cepstrum and the first-order differential Mel frequency cepstrum coefficient. At the same time, an improved speech signal feature parameter extraction algorithm is proposed. The calculation steps of linear predictive cepstrum coefficients based on wavelet transform and Mel frequency cepstrum coefficients based on wavelet transform are presented. Finally, the coefficient matrix of linear predictive cepstrum coefficient (DWTL) based on wavelet transform and corresponding difference parameter (DWTL), Mell-frequency cepstrum coefficient (DWTM) based on wavelet transform and the corresponding difference parameter (DWTM) are used. Then the model training is realized by the Baum-welch algorithm, which is a forward and backward algorithm, which is used in the hidden Markov model. At the same time, the interactive interface of speech recognition simulation is realized by the design of Matlab GUI and the writing of callback function. On the basis of speech recognition theory, combined with Arduino dual-arm robot. The forward kinematics problem and inverse kinematics problem of the robot arm are solved by the coordinate general rotation transformation algorithm. The forward kinematics problem is to solve the pose of the end actuator by the known joint variables, and the inverse kinematics problem can obtain the rotation angle of each joint according to the position and attitude requirements of the robot end actuator. Then, the ASR M08-A speech recognition module is used to control the robot arm with 5 degrees of freedom by using the 32 path steering gear control board and the Arduino atmegal2560 control board.
【學(xué)位授予單位】：廣東工業(yè)大學(xué)
【學(xué)位級別】：碩士
【學(xué)位授予年份】：2014
【分類號】：TP242;TN912.34
本文編號：2208813
[Abstract]:This paper mainly studies the whole process of speech recognition: it consists of speech signal filtering, sampling, quantization, windowing, endpoint detection, feature extraction, model training and threshold comparison, and the simulation of the algorithm model through Matlab. At the same time, the interactive interface of speech recognition is realized by GUI design technology of Matlab. On the basis of speech recognition theory, a five-degree-of-freedom Arduino dual-arm robot and a ASR M08-A speech recognition module are built to realize various planning actions of the speech control robot. After the speech signal is filtered, sampled and quantized, the discrete digital signal is pre-accentuated. The purpose of preemphasis is to filter out the low-frequency interference and enhance the high frequency component of the input signal. Framing makes the original signal a segment, which is equivalent to adding a rectangular window to the original signal. Multiplying with the rectangular window in the time domain is equivalent to convolution between the frequency spectrum of the signal and the Fourier transform of the rectangular window. The dual threshold endpoint detection algorithm realizes the endpoint detection of speech signal through two thresholds: the short time average energy and the zero crossing rate. After the endpoint detection of speech signal is realized, the characteristic parameters of speech signal are obtained by the number of Mel frequency cepstrum and the first-order differential Mel frequency cepstrum coefficient. At the same time, an improved speech signal feature parameter extraction algorithm is proposed. The calculation steps of linear predictive cepstrum coefficients based on wavelet transform and Mel frequency cepstrum coefficients based on wavelet transform are presented. Finally, the coefficient matrix of linear predictive cepstrum coefficient (DWTL) based on wavelet transform and corresponding difference parameter (DWTL), Mell-frequency cepstrum coefficient (DWTM) based on wavelet transform and the corresponding difference parameter (DWTM) are used. Then the model training is realized by the Baum-welch algorithm, which is a forward and backward algorithm, which is used in the hidden Markov model. At the same time, the interactive interface of speech recognition simulation is realized by the design of Matlab GUI and the writing of callback function. On the basis of speech recognition theory, combined with Arduino dual-arm robot. The forward kinematics problem and inverse kinematics problem of the robot arm are solved by the coordinate general rotation transformation algorithm. The forward kinematics problem is to solve the pose of the end actuator by the known joint variables, and the inverse kinematics problem can obtain the rotation angle of each joint according to the position and attitude requirements of the robot end actuator. Then, the ASR M08-A speech recognition module is used to control the robot arm with 5 degrees of freedom by using the 32 path steering gear control board and the Arduino atmegal2560 control board.
【學(xué)位授予單位】：廣東工業(yè)大學(xué)
【學(xué)位級別】：碩士
【學(xué)位授予年份】：2014
【分類號】：TP242;TN912.34
【引證文獻(xiàn)】
相關(guān)期刊論文 前1條
1 左軒塵;韓亮亮;莊杰;石琪琦;黃煒;;基于ROS的空間機(jī)器人人機(jī)交互系統(tǒng)設(shè)計[J];計算機(jī)工程與設(shè)計;2015年12期
相關(guān)碩士學(xué)位論文 前1條
1 朱健晨;基于語音信號特征參數(shù)提取的同模板匹配算法的綜合分析及應(yīng)用[D];昆明理工大學(xué);2015年
，本文編號：2208813
本文鏈接：http://sikaile.net/kejilunwen/wltx/2208813.html
最近更新
教材專著
