【作者】 梁文彬；

【导师】 张帆；

【作者基本信息】 湖南大学 ， 控制理论控制工程， 2008， 硕士

【摘要】 音乐是一门艺术,音乐识别是语音识别的一个分支,是科学和艺术的交叉。由于音乐的专业性强、乐理知识复杂、音乐变化多样等因素,专业的基于嵌入式的音乐识别产品至今很少而且不完善,在音乐识别这个新领域,由于其环境和行业的特殊性,需要专门进行开发,以适应社会需要。本文设计并实现了一种基于TMS320VC5402DSP的嵌入式音乐识别系统。详细阐述了嵌入式音乐识别算法的详细设计与实现,研究了音乐语言的特点和音乐识别的侧重点,并阐述了基于音乐信号特色的语音处理及识别方法;针对音乐信号中的毛刺干扰,提出了曲线整形的思想,消除了音乐信号中的毛刺;针对音乐信号端点检测困难的特点,利用多频段能量曲线分割结合过零率来实现端点准确检测;针对音高提取运算量大且容易受共振峰影响的特点,利用线性预测残差的方法对传统的AMDF算法进行了改进并提取出了音高,在实际实现时,优化了AMDF算法,减少了计算量;为了得到较好的特征参数,通过求取梅尔频标倒谱系数的方法,分别提取了模板信号特征参数和待测信号特征参数;为了提高识别率,针对传统DTW算法的缺陷,采用了放宽端点和声刺激法改进了传统的DTW算法的性能并进行了仿真试验,使音乐识别率得到明显提高;在音高、节奏评分时,针对对位评分的缺点,采取了动态调整的方法提高了评分的准确度。在硬件实现上,详细阐述了基于TMS320VC5402DSP的嵌入式音乐识别系统的各部分硬件设计,在软件开发上,给出了嵌入式音乐识别系统软件设计的各部分流程,并对各部分进行了仿真试验,给出了仿真结果,并分别对声乐和器乐这两种音乐信号进行了模板特征参数提取及其与待测信号进行匹配识别,试验结果表明:对音乐信号的识别精度在96%以上,成功实现了对音乐的音高、唱名、节奏的识别,满足了实际应用的需求。更多还原

【Abstract】 Music is an art. Music recognition is an embranchment of the speech recognition, it is a crossing of the science and the art. Because the music signal is very special,it is a ramdon signal, and the music theory is very complex, the products based on the embedded music recognition is a fat lot and with faultiness. In this new field,because of it’s particularity,often need specially developing,in order to meet the needs of the society.In this paper, we designed a music recognition system based on TMS320VC5402 DSP, and make it come true. Firstly, we particularly explained the algorithmic of the music recognition. Then, we made a study of the music characteristic and the music recognition methods. As for the burrs of the music signal, we put forward a curve plastic method to eliminate the burrs. Considering the difficulties of detecting the point of the music speech, we made use of the multi-frequency energy curve to detect the end-point of the music signal with the crossing zeros rate. As for the disadvantages of the large operation and influence of the formant, we improved the AMDF theory with the linear prediction algorithmic to detect the error. In order to obtain the characteristic parameter of the music signal,wo calculate the MFCC parameter to obtain the template and the pending signal characteristic parameter. In order to improve the rate of the recognition, considering the disadvantages of the DTW algorithmic, we find a new method to improved the DTW algorithmic by broadening the point of the music signal and made simulation experimentation; As for the disadvantages of the contraposition grade to the pitch, cadent and sing name, we find a method to dynaic adjust the results of the recognition to improve the accuracy of the grade.As for the realization of hardware, the thesis depicts the realization of every part of music recognition system based on the TMS320VC5402 in detail; as for the development of software, the thesis gives the software design flow chart of the music recognition system, simulates the basic theory with MATLAB language and gives the simulation results. Well, with the characteristic parameters of the vocality and the instrumental music, we can recognize the input digit music speech successfully and put forward it’s pitch, cadent, and sing name. The result is comform that the music recognition based on the TMS320VC5402 is run well and the accuracy reaches 96 percent.and it can meet the needs of the practicality application.更多还原