Research On Audio Frame Erasure Concealment Method For ACELP-based Codec

In this paper, we proposed a new audio frame erasure concealment method for ACELP-based codec, which could mitigate the adverse influence of packet lost or audio frame erasure on the reconstructed speech quality of decoder. At first, the procedure of constraining the energy contribution of adaptive codebook to the synthesized speech signal was embedded into the encoder to speed up the recovery and prevent the error propagation.Its only drawback was minor degradation in quality in error-free channel compared to the original codec at the same bit rate. At the decoder, we needed to determine the speech type of the last previous frame and the first future frame before concealment, and then decided the speech type for the erased frame. The speech type could be classified as voiced, unvoiced and transitions. Then applied two different mathematic models to reconstruct the ACELP parameters: the Lagrange interpolation polynomial approximation formula was used to restore the ISP parameters or LSF parameters and a linear prediction model was applied to recover the pitch and gains of the innovative codebook. The parameters in the erased frames were reconstructed respectively. The proposed method made use of not only the previous valid frames but also the future frames to rebuild the parameters of ACELP model. Compared with the traditional concealment methods which only used the past valid frames, the proposed modification at the decoder could restore some most important encoder parameters more accurate and efficient.We applied the three technologies above to the ITU-T AMR-WB codec, and meanwhile implemented the concealment method of ITU-T G.729 codec on the AMR-WB decoder. The waveforms in Time-field, Mean Opinion Score, Perceptual Evaluation of Speech Quality and Objective Difference Grade were used to compare the output speech quality of the proposed technique with the existing methods. The experimental results demonstrated that the proposed concealment method achieved obvious performance improvement over the standard codec and G.729 concealment method for a variety of speech sequences and various frame erasure rates ranging from 1% to 10%.The proposed method could be applied to virtually any ACELP-like codec, however, in our research, we focused on their application to the AMR-WB standard. The proposed method required neither extra data nor too much additional delay (no more than 20ms) and was completely interoperable with the standard codec. With its high fidelity, low delay and low complexity, therefore, it can be readily applied to various consumer electronic applications.