Content - Based Design Of Non - Equal Protection Music Packet Loss Recovery

Bandwidth efficiency and error robustness are two essential and conflicting requirements forstreaming media content over error-prone channels, such as wireless channels. Most publishedworks in audio packet loss recovery have focused on speech. Relatively few published works havedealt with high quality music streaming. Because of their different purposes, their qualityrequirements in streaming applications are quite different. Therefore, the technologies which areapplied to the audio packet loss do not suit to the music streaming. These technologies do not usethe features of the music. As a result, we proposed a novel scheme proposed "content-based unequalerror protection". The main contents of this thesis are as follows:1.For the transients, which exist in the music streaming, we processed sub-frame and extractthe data frames with the onset point. Through this process, we determine the position of the onsetby calculating the temporal energy/harmonic structure/zero crossing rate/phase change.2.For the different characteristics of the onset point, we use the SVM classification. We use thedispenser wrapper mode, according to the features-the temporal energy/harmonicstructure/bandwidth; we form a feature vector to classify the feature space into two classes (Noteonset and Drumbeat).3.According to the characteristics of different types of onset point, we do some recoveryrespectively. Ultimately, we will send the secondary bit stream to the receiver side. We can recoverthe packet loss by the secondary bit stream.Formal perceptual evaluation shows that our scheme improves audio quality significantly oversimple muting and packet repetition. This improvement is achieved with a negligible amount ofredundancy, which is transmitted to the receiver ahead of playback.