Research On Technologies Of Packet Loss Concealment For Mobile Audio Coding

With the rapid development of mobile internet, a large amount of demands for audio and musical applications increase on mobile terminals. However, considering the no-QoS factor of wireless and mobile networks as well as the rapidly varied wireless channel, it leads to the bit error or distortion of signal data obtained by receiver and results in poor quality of voice transmission and audio communication service.Nowadays, packet loss concealment technology for mobile audio transmission research mainly on problems of small random lost voice packets in audio application scenarios,but it can not effectively deal with unexpected large number of continuous packet loss scenarios. The paper carries out related research to solve the problem.In the paper, based on a full study of the mobile audio coding standard AMR-WB+ and AVS-P10, with deep discussion of packet loss concealment methods of mobile audio coding for mainstream ACELP encoder and TCX encoder, some novel packet loss concealment schemes are addressed to meet the demand for high recovery efficiency and low-latency audio transmission for mobile network,so that the encoding side does not add additional error correction packets to the audio stream and the decoding side does not increase additional delay. The main research work contains the following three aspects:1. An adaptive packet loss concealment scheme for ACELP codecPitch and ISF are key parameters for ACELP codec. Current packet loss concealment scheme in AVS-P10 standard is suitable for single packet or small amount of lost packets, but not for large amounts of continuous and burst of packet loss with higher delay and poor quality of audio recovery. According to this, a new adaptive packet loss concealment scheme called A-PLC is proposed. The scheme utilizes different packet loss concealment methods to recover pitch and ISF parameters according to single packet loss or multi-packet continuous loss respectively. Performance comparison is conducted between the A-PLC scheme and the original scheme employed in AVS-P10 standard. The experiment results show that the test objective PESQ score achieves an increase of 0.13 points, and subjective CMOS auditory test score achieves an increase of 0.04 points. The computation time achieves a decrease of 10%.2. An HMM-based packet loss concealment scheme for ACELP codecThe packet loss concealment methods in most of the current audio codecs are mainly based on the estimation of packet loss parameters by the replacement of correct parameter information before or after the lost packet or linear prediction. The main drawback of current methods only focuses on the local features of audio signal, but not the global features of the whole audio signal. This leads to the larger distortion under the circumstance of the burst of large amount of packet loss. Base on this reason, an HMM-based packet loss concealment scheme called H-PLC is proposed. The scheme utilizes HMM model to decide the appropriate packet loss concealment strategy according to the statistical changes of larger range of context in audio signal. The codec parameter vector in each packet is as the observation value in HMM model. A series of parameter vectors are as the continuous probability density function. While observing the change of observed value in audio parameters through HMM model in codec, it needs to decide the most possible signal status at each moment. When packet loss occurs, the estimation is involved by the recovery through status and probability density function. The experiment results show that the objective listening test PESQ score achieves an increase of 0.33 points, subjective listening test MOS score achieves an increase of 0.32 points when compared with the original scheme employed in AVS-P10 standard.3. A packet loss concealment scheme based on gain control for TCX codecThe current packet loss concealment methods mostly use the amplitude gain to control the magnitude of the recovery of the spectral coefficients, which cannot reflect the differences of the amplitude changes of the spectral coefficients. This leads to the larger distortion of recovered audio signal. Therefore, a packet loss concealment scheme based on gain control called Gain-C is proposed. The scheme utilizes the distance of line spectrum frequency coefficiency to represent the stability of filter. Based on this, the scheme builds the linear transformation relationship between the gain control factor of the amplitude and the stability factor so as to adaptively control the recovery of spectral coefficiency. The experiment results show that the recovered audio signal achieves an increase of 0.05 dB of the average weighted SNR and an increase of 1.5 points of MUSHRA listening test average score when compared with original scheme employed in AMR-WB+ standard.The addressed packet loss concealment schemes can be coded and directly embedded in the relative processing section in mobile audio decoder. Research results show that:the quality and efficiency performance of the proposed packet loss concealment schemes are superior to current mainstream packet loss concealment methods employed in AMR-WB+ and AVS-P10 standard.