No-Reference Objective Quality Assessment For Networked Audio

Audio quality assessment is the key technology to ensure the quality of networked audio services, with the rapid development of multimedia and the wide application of audio products, the objective audio quality assessment is becoming increasingly important. In order to provide the reliable quality of the audio for the end user, it is usually required to monitor the audio quality in the network nodes, and then adjust the coding and channel parameters to guaranteeing the audio quality. Since it is difficult to obtain the original signal as a reference in the network environment, the no-reference audio quality assessment method shows a high application value and is very suitable for monitoring and evaluating the quality of the networked audio.Considering the characteristics of networked speech communication, the paper presents a new method for evaluating the quality of speech based on signal. The method parses the data packets header information and payload information, and the speech of payload are decoded to obtain the signal waveform, through the introduction of human auditory perception model, analyze the characteristics of the speech signal time-domain spectral envelope, extract feature parameters can reflect the speech quality to predict the quality distortion caused by compression and background noise factors. In addition, and in the process of decoding packets extract speech packet loss rate parameters to predict the quality distortion caused by the factors of packet loss. Finally, the above three kinds of distortion factors are used to obtain the final speech perception quality.Audio compression is an important factor influencing the audio quality. In order to evaluate the impact of audio compression on the audio quality, taking the common networked audio AAC as an example, a novel method for the evaluation of the compression distortion based on the parameters of the AAC encoding audio is presented. In AAC encoding algorithm, the encoding bit-rate and quantization factor are two main factors causing the distortion of audio compression, therefore, the related parameters are extracted to estimate the compression distortion from the decoding process. The proposed method extracts the quantization factor which can reflect the complex characteristic of the audio content, and incorporates it into the model between the encoding bit-rate and compression distortion, thus the proposed method can effectively evaluate the distortion of the networked audio with different contents.