Study On Audio Error Concealment Algorithm Based On Waveform Reconstruction

People living in the environment of a variety of sounds, language conveys information in social communication activities and the music expresses the feelings of the people. When the compressed audio signal in the process of storage and transmission is lost, the use of audio error concealment (EC) to deal with the loss of audio signals is necessary. This technique uses short-term stable characteristics of the audio signal, combined with characteristics of human auditory system to cover up audio compression data which has some decoding errors caused by damaged storage media and transmission channel errors in order to improve the audio playback quality. So far, many scholars put forward the audio error concealment algorithms, which are divided into two categories, one is sender-based repair and the other is receiver-based repair. Each of these algorithms has their own advantages and disadvantages.The study in this paper is primarily aimed at proposing an audio error concealment algorithm based on waveform reconstruction. In this ideology, derived from three types of implementations of the program, which are audio error concealment based on curve fitting, audio error concealment algorithm based on pattern matching and short-time zero-crossing detection and audio error concealment based on wavelet decomposition and reconstruction. This thesis is full use of the correlation between the audio frames. When the audio frames are lost, one of the three proposed implementation schemes is used to cover the lost frames. The ultimate goal is to maximize the recovery the waveform of the lost signal and to complete the reconstruction of the audio signal.The main task of this thesis is summarized as follows:First, based on the idea of reconstructing the waveform of the lost signal, this paper proposes the first implementation, which is the audio error concealment based on curve fitting. This method takes advantage of the fitting model and the correlation between the audio frames to fit the waveform of the lost signal. This approach directly deals with the audio signal sample values, so it is superior to other approaches in the wider range of applications. This is done by fitting the right frame before and after the lost frame respectively to obtain the fitting coefficients, and then computes the weighted average of the coefficients as the coefficients of the missing frame. At last the polynomial values are reversed to recover the lost frame. This approach improves the quality of the audio signal compared with other audio error concealment algorithms. But increase of the fit factors will bring amplitude fluctuations to the audio signal and the recovery of the audio signal would be affected, so the following two implementations are proposed. Second, for the impact of the methods has been described above on the waveform, here is presented directly from the audio signal waveform characteristics of the time domain, looking for processing methods to cover the lost signal. That is adding the short-time zero-crossing detection module to the existed pattern matching method. The short-time zero-crossing can clearly depicts the development trend of the audio signal waveform. When the original audio irregular or the correlation is not strong between signals, the loss can not be reconstructed very well by the original pattern matching method, so it can be overcome the shortcomings by the combination of this method. Through a large number of experiments found that compared with the existing audio error concealment methods the objective evaluation criteria of SNR is increased and the recovery of the waveform is also closer to the original signal by this approach. And compared with the first implementation of this paper, instability is improved by this method.Finally, for the first implementation of the instability and the second implementation that due to the limited length of the window the matching search result is not strong similarity to the lost signal, so the third implementation which is based on wavelet decomposition and reconstruction is proposed. This implementation is to do wavelet decomposition with the correct frames before and after the lost frame and to obtain coefficients from wavelet decomposition, and then these two sets of wavelet coefficients are processed as the wavelet coefficients for the lost frame and recover the lost frame by the wavelet reconstruction. Treatment of coefficients in two ways, one is to do a whole average to the coefficient vector; the other is the coefficient refinement, that is, the adjacent frames around the lost frame are decomposed by wavelet and wavelet coefficients obtained by the maximum layer search for relevance, the final combination of the wavelet coefficients as the coefficients of the lost frames, according to the scope of the search processing is divided into unilateral and bilateral treatment. According to a large number of experimental results, this implementation increases the audio quality of restoration compared to the original method and the two implementations proposed by this paper described before.In summary, the algorithms in this thesis have achieved the purpose to reconstruct the waveform of the missing frame waveform. In order to make the proposed algorithm has better versatility and get higher quality audio restoration, finding the appropriate fitting model, choosing a more reasonable criterion of matching criteria and the influence on audio restoration by the different layers of the wavelet coefficients can be further refinement, these will be the focus of future research work.