基于非对等保护的音乐包丢失恢复的研究与实现

发布时间：2018-01-10 16:02

本文关键词：基于非对等保护的音乐包丢失恢复的研究与实现　出处：《南京邮电大学》2014年硕士论文　论文类型：学位论文

【摘要】：在传输数据易发生错误的信道(例如无线信道)中进行数据传输时,频带利用率和抗错鲁棒性是传输系统必须考虑且相互冲突的两个部分。现阶段已经有很多关于语音信号包丢失的恢复方法。然而,其中仅仅只有很少的方法可用于处理恢复出高质量的音乐数据。音乐信号与一般的语音信号相比,二者的目的不同、应用中用户的感知质量要求不一,且当前绝大部分的音频包丢失恢复技术并没有充分的利用到音乐内容本身的一些特性。所以本文在当前的一些包丢失恢复技术的基础上提出了“基于非对等保护”的音乐包丢失恢复方法。本文的主要工作如下：(1)针对音乐信号本身存在的瞬变现象。本文对音乐信号进行了分帧处理、提取含有onset点的数据帧。该过程中本文主要通过信号的短时能量、谐波结构、过零率、相位变化等特征向量来确定onset点的时域位置。(2)利用onset点的不同特征来处理onset点的分类问题。本文采用了SVM分类方法,采用分装器wrapper模式。根据不同onset点的短时能量曲线、频谱谐波结构、频带宽度建立特征向量,从而将onset点分为Note onset及Drumbeat两类瞬变点。(3)根据不同的类型的onset点的特性,本文分别对含有Note onset或Drumbeat的丢帧进行恢复。并将每一帧的特征报文信息在缓冲的时候发送给接收端。以便于接收端对丢失的帧进行恢复。我们的实验结果显示本文的“基于非对等保护”的音乐包丢失恢复方法相比较于当下的简单的噪声抑制、包替代恢复方法在处理音乐数据流方面,有着更为显著的效果。而且本文所提出的方法相对于当前的方法而言,仅仅只需要很少的冗余信息就能较好的恢复出音乐信号。
[Abstract]:When data is transmitted in a channel, such as a wireless channel, where the transmission data is prone to errors. Frequency band efficiency and anti-error robustness are two parts of transmission system which must be considered and conflicting. There are many restoration methods for voice packet loss at this stage. However. Only a few of the methods can be used to process and restore the high quality music data. Compared with the normal speech signal, the purpose of the two methods is different, and the user's perception quality requirements are different in the application. And most of the current audio packet loss recovery techniques have not fully utilized some of the characteristics of the music content itself. Music packet loss recovery method based on non-peer-to-peer protection. The main work of this paper is as follows:. 1) aiming at the transient phenomenon of the music signal itself, this paper carries on the frame dividing processing to the music signal. Data frames with onset points are extracted. In this process, the short time energy, harmonic structure and zero crossing rate of the signal are mainly used in this paper. Phase change and other Eigenvectors are used to determine the time domain position of onset points. (2) the different features of onset points are used to deal with the classification of onset points. SVM classification method is used in this paper. According to the short time energy curve of different onset points, the structure of harmonic spectrum and the frequency band width, the eigenvector is established by using the splitter wrapper mode. Thus, the onset points are divided into Note onset and Drumbeat transients. 3) according to the characteristics of different types of onset points. This article respectively contains Note. The lost frame of onset or Drumbeat is restored, and the feature message of each frame is sent to the receiver when buffering. The result of our experiment is that the receiver can recover the lost frame. Shows the. The music packet loss recovery method based on non-peer-to-peer protection is compared with the current simple noise suppression method. Packet substitution recovery method has a more significant effect on the processing of music data flow, and the method proposed in this paper is more effective than the current method. Only a small amount of redundant information is needed to recover the music signal.
【学位授予单位】：南京邮电大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN912.3

【相似文献】