Packet Loss Concealment in Networked Music Performance using Auto-Regressive Models

RTP implementation of a packet loss concelament method for networked music performance: A Networked Music Performance (NMP) is a real-time Internet application, in which different performers or musicians hold the music show remotely and interact with each other going through communication networks. For enabling musicians play simultaneously in distance, a low latency transmission is strictly required. Therefore, to minimize the end-to-end delay, uncompressed, bidirectional audio streams are used in some typical implementations of NMP applications, along with UDP running at transport layer. However, due to the best effort mechanism of the Internet, it is hard to avoid the network jitter and data loss. What's more, UDP does not support acknowledgement and re-transmission mechannism to recover the lost or excessive delayed data. In order to make NMP realistic with a high quality for musicians and audience, exploring packet loss concealment approaches (PLC) of NMP to solve this problem is becoming a more and more interesting topic in the last decades for musicians and engineers. In this thesis, we propose a PLC method relying on the state-of-the-art Auto-Regressive Models, which was developed by statisticians firstly to do time series forecasting. We use this technique to predict the lost music data and analyse the performance by adjusting the related parameters for different genres of music.