本文選題：語音識(shí)別　切入點(diǎn)：非特定人　出處：《安徽工業(yè)大學(xué)》2016年碩士論文

【摘要】：在科學(xué)技術(shù)發(fā)展的推動(dòng)下,語音識(shí)別技術(shù)已經(jīng)逐漸從研究階段進(jìn)入到實(shí)際應(yīng)用階段。但是,對(duì)非特定人的語音識(shí)別研究仍在激烈的探討中,怎樣提高該系統(tǒng)的識(shí)別率,依舊是當(dāng)前研究的熱點(diǎn)問題。本文系統(tǒng)地研究了語音識(shí)別系統(tǒng)的各個(gè)組成部分,針對(duì)部分關(guān)鍵技術(shù)提出了改進(jìn)的算法,并在MATLAB上建立了相應(yīng)的非特定人識(shí)別系統(tǒng)。文中深入研究了語音識(shí)別系統(tǒng)的原理組成部分,包括語音信號(hào)的預(yù)處理、起止端點(diǎn)的檢測(cè)、特征參數(shù)的提取。在此基礎(chǔ)上,對(duì)三種常用的語音識(shí)別方法:動(dòng)態(tài)時(shí)間規(guī)整(DTW)、隱馬爾科夫模型(HMM)與神經(jīng)網(wǎng)絡(luò)模型(ANN)進(jìn)行了對(duì)比分析,并重點(diǎn)研究了隱馬爾科夫模型算法,對(duì)該算法中存在的數(shù)據(jù)溢出問題采取了有效的解決措施。接著,針對(duì)低信噪比噪聲環(huán)境下,語音信號(hào)的濾波和端點(diǎn)檢測(cè)這兩個(gè)關(guān)鍵技術(shù),分別提出了改進(jìn)的算法,即:基于經(jīng)驗(yàn)?zāi)Ｊ椒纸?EMD)和奇異值分解(SVD)差熵法的濾波算法,以及改進(jìn)的希爾伯特黃變換(HHT)語音端點(diǎn)檢測(cè)算法,并將改進(jìn)后的算法分別與傳統(tǒng)算法的處理結(jié)果進(jìn)行了分析比較。本文在MATLAB平臺(tái)上建立了基于HMM模型的非特定人的語音識(shí)別系統(tǒng)。結(jié)果表明,與傳統(tǒng)的濾波方法以及端點(diǎn)檢測(cè)方法相比,改進(jìn)后的算法提高了識(shí)別系統(tǒng)的識(shí)別率,充分體現(xiàn)了改進(jìn)算法的有效性和可行性。最后設(shè)計(jì)了一個(gè)語音識(shí)別系統(tǒng)GUI界面,包括語音信號(hào)處理的界面和語音的識(shí)別過程界面,對(duì)語音庫(kù)中的語音進(jìn)行實(shí)時(shí)識(shí)別實(shí)驗(yàn),驗(yàn)證了所用系列方法的有效性。
[Abstract]:With the development of science and technology, speech recognition technology has gradually moved from the research stage to the practical application stage. However, the research on the speech recognition of non-specific people is still under intense discussion, how to improve the recognition rate of the system, It is still a hot topic in current research. In this paper, the components of speech recognition system are systematically studied, and an improved algorithm is proposed for some key technologies. In this paper, the principle of speech recognition system is deeply studied, including speech signal preprocessing, endpoint detection and feature parameter extraction. Three common speech recognition methods: dynamic time warping (DTW), Hidden Markov Model (HMMM) and Neural Network (Ann) are compared and analyzed. This paper takes effective measures to solve the problem of data overflow in the algorithm. Then, aiming at the two key technologies of speech signal filtering and endpoint detection in low signal-to-noise noise environment, the improved algorithm is proposed respectively. That is, the filtering algorithm based on empirical mode decomposition (EMD) and singular value decomposition (SVD) differential entropy method, and the improved Hilbert Huang transform (HHT) speech endpoint detection algorithm. The improved algorithm is analyzed and compared with the results of the traditional algorithm. In this paper, a speech recognition system based on the HMM model is established on the MATLAB platform. The results show that, Compared with the traditional filtering method and the endpoint detection method, the improved algorithm improves the recognition rate of the recognition system, and fully reflects the effectiveness and feasibility of the improved algorithm. Finally, a speech recognition system GUI interface is designed. It includes the interface of speech signal processing and the interface of speech recognition process. The experiments of speech recognition in speech database are carried out in real time, and the validity of the series of methods is verified.
【學(xué)位授予單位】：安徽工業(yè)大學(xué)
【學(xué)位級(jí)別】：碩士
【學(xué)位授予年份】：2016
【分類號(hào)】：TN912.34

【相似文獻(xiàn)】

相關(guān)期刊論文 前10條

1 汪洪波;;語音識(shí)別系統(tǒng)在配送中心的應(yīng)用[J];信息與電腦;2006年06期

2 楊q，

本文編號(hào)：1669740