No-reference Objective Quality Assessment For EVS Audio Streams

With the development of the network,the network audio services have been widely used.The development of network audio services is inseparable from the support of the encoder.The fourth generation of new audio codec EVS provided by 3GPP with its excellent performance has been widely used in different scenarios.The audio quality provided by the EVS codec,however,can be influenced by many factors in the practical application scenarios.To ensure the user experience,it is needed to evaluate the audio quality of EVS and adjusts the coding and network parameters according to the evaluation results to provide a reliable audio quality to the end user.Therefore,it is quite meaningful to study the audio quality provided by the EVS coding standard.In order to analyze the influence of various factors on voice quality during the network audio compression and transmission,a network audio real-time transmission system is constructed,where multiple functions,such as the audio signal encoding,network transmission and decoding,can be served.It can also set different coding parameters and network parameters.This system can be applied to the furhter analysis of the distortion factors that affect the quality of network audio.Focusing on the influence of the background noise,coding compression and network packet loss on the quality of the voice over IP,this thesis presents an EVS speech quality evaluation model considering background noise.Firstly,the speech signal is obtained by decoding the IP packet.The time domain envelope of the obtained speech signal is further analyzed.The spectral characteristics of the envelope are extracted and utilized to estimate the speech quality affected by the background noise.Secondly,analyzing the influence of the coding rate and the speech quality affected by the background noise on the coding quality.Finally,based on the obtained coding quality,the final speech quality is predicted by evaluating the influence of coding quality and the packet loss rate on the speech quality.The experimental results show that the EVS speech quality evaluation model proposed in this thesis can accurately predict the EVS speech quality.In order to analyze the influence of different types of audio(speech and music)on the coding quality,a packet audio coding distortion evaluation model is proposed based on the classification of audio types.The proposed model calculates the coding rate of the packets by analyzing the packet header information,firstly.Then,the audio type of the packet is classified by detecting the packet size.Finally,analyzing the influence of the coding rate and the different types of audio(speech and music)on the coding quality to estimate the coding quality of the EVS audio stream.The experimental results show that the coding quality evaluated by the method proposed in this thesis have a good consistency with the subjective quality.